diff --git a/.drone.yml b/.drone.yml
index 696c5a99d..b1c211d14 100644
--- a/.drone.yml
+++ b/.drone.yml
@@ -8,7 +8,7 @@ platform:
 
 steps:
 - name: Build and Test
-  image: ubuntu:19.04
+  image: ubuntu:18.04
   environment:
     CC: gcc
     COMMON_FLAGS: 'DYNAMIC_ARCH=1 TARGET=ARMV8 NUM_THREADS=32'
@@ -32,7 +32,7 @@ platform:
 
 steps:
 - name: Build and Test
-  image: ubuntu:19.04
+  image: ubuntu:18.04
   environment:
     CC: gcc
     COMMON_FLAGS: 'DYNAMIC_ARCH=1 TARGET=ARMV6 NUM_THREADS=32'
@@ -141,3 +141,52 @@ steps:
     - cmake $CMAKE_FLAGS ..
     - make -j
     - ctest -V
+
+---
+kind: pipeline
+name: arm64_native_test
+
+platform:
+  os: linux
+  arch: arm64
+
+steps:
+- name: Build and Test
+  image: ubuntu:18.04
+  environment:
+    CC: gcc
+    COMMON_FLAGS: 'USE_OPENMP=1'
+  commands:
+    - echo "MAKE_FLAGS:= $COMMON_FLAGS"
+    - apt-get update -y
+    - apt-get install -y make $CC gfortran perl python g++
+    - $CC --version
+    - make QUIET_MAKE=1 $COMMON_FLAGS
+    - make -C test $COMMON_FLAGS
+    - make -C ctest $COMMON_FLAGS
+    - make -C utest $COMMON_FLAGS
+    - make -C cpp_thread_test dgemm_tester
+---
+kind: pipeline
+name: epyc_native_test
+
+platform:
+  os: linux
+  arch: amd64
+
+steps:
+- name: Build and Test
+  image: ubuntu:18.04
+  environment:
+    CC: gcc
+    COMMON_FLAGS: 'USE_OPENMP=1'
+  commands:
+    - echo "MAKE_FLAGS:= $COMMON_FLAGS"
+    - apt-get update -y
+    - apt-get install -y make $CC gfortran perl python g++
+    - $CC --version
+    - make QUIET_MAKE=1 $COMMON_FLAGS
+    - make -C test $COMMON_FLAGS
+    - make -C ctest $COMMON_FLAGS
+    - make -C utest $COMMON_FLAGS
+    - make -C cpp_thread_test dgemm_tester
diff --git a/.github/workflows/dynamic_arch.yml b/.github/workflows/dynamic_arch.yml
new file mode 100644
index 000000000..b6a4090bd
--- /dev/null
+++ b/.github/workflows/dynamic_arch.yml
@@ -0,0 +1,81 @@
+name: continuous build
+
+on: [push, pull_request]
+
+jobs:
+  build:
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [ubuntu-latest, macos-latest]
+        build: [cmake, make]
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v2
+
+      - name: Compilation cache
+        uses: actions/cache@v2
+        with:
+          path: ~/.ccache
+          # We include the commit sha in the cache key, as new cache entries are
+          # only created if there is no existing entry for the key yet.
+          key: ${{ runner.os }}-ccache-${{ github.sha }}
+          # Restore any ccache cache entry, if none for
+          # ${{ runner.os }}-ccache-${{ github.sha }} exists
+          restore-keys: |
+            ${{ runner.os }}-ccache
+
+      - name: Print system information
+        run: |
+          if [ "$RUNNER_OS" == "Linux" ]; then
+            cat /proc/cpuinfo
+          elif [ "$RUNNER_OS" == "macOS" ]; then
+            sysctl -a | grep machdep.cpu
+          else
+            echo "$RUNNER_OS not supported"
+            exit 1
+          fi
+
+      - name: Install Dependencies
+        run: |
+          if [ "$RUNNER_OS" == "Linux" ]; then
+            sudo apt-get install -y gfortran cmake ccache
+          elif [ "$RUNNER_OS" == "macOS" ]; then
+            brew install coreutils cmake ccache
+          else
+            echo "$RUNNER_OS not supported"
+            exit 1
+          fi
+          ccache -M 300M  # Limit the ccache size; Github's overall cache limit is 5GB
+
+      - name: Build
+        if: matrix.build == 'make'
+        run: |
+          if [ "$RUNNER_OS" == "Linux" ]; then
+            export PATH="/usr/lib/ccache:${PATH}"
+          elif [ "$RUNNER_OS" == "macOS" ]; then
+            export PATH="$(brew --prefix)/opt/ccache/libexec:${PATH}"
+          else
+            echo "$RUNNER_OS not supported"
+            exit 1
+          fi
+
+          make -j$(nproc) DYNAMIC_ARCH=1 USE_OPENMP=0
+
+      - name: CMake build
+        if: matrix.build == 'cmake'
+        run: |
+          if [ "$RUNNER_OS" == "Linux" ]; then
+            export PATH="/usr/lib/ccache:${PATH}"
+          elif [ "$RUNNER_OS" == "macOS" ]; then
+            export PATH="$(brew --prefix)/opt/ccache/libexec:${PATH}"
+          else
+            echo "$RUNNER_OS not supported"
+            exit 1
+          fi
+
+          mkdir build
+          cd build
+          cmake -DDYNAMIC_ARCH=1 -DNOFORTRAN=0 -DBUILD_WITHOUT_LAPACK=0  -DCMAKE_VERBOSE_MAKEFILE=ON -DCMAKE_BUILD_TYPE=Release ..
+          make -j$(nproc)
diff --git a/.github/workflows/nightly-Homebrew-build.yml b/.github/workflows/nightly-Homebrew-build.yml
index f55e73d23..8d7cfea2d 100644
--- a/.github/workflows/nightly-Homebrew-build.yml
+++ b/.github/workflows/nightly-Homebrew-build.yml
@@ -21,6 +21,7 @@ jobs:
   build-OpenBLAS-with-Homebrew:
     runs-on: macos-latest
     env:
+      DEVELOPER_DIR: /Applications/Xcode_11.4.1.app/Contents/Developer
       HOMEBREW_DEVELOPER: "ON"
       HOMEBREW_DISPLAY_INSTALL_TIMES: "ON"
       HOMEBREW_NO_ANALYTICS: "ON"
diff --git a/.gitignore b/.gitignore
index 6803a919e..bca79f043 100644
--- a/.gitignore
+++ b/.gitignore
@@ -70,6 +70,7 @@ test/SBLAT2.SUMM
 test/SBLAT3.SUMM
 test/ZBLAT2.SUMM
 test/ZBLAT3.SUMM
+test/SHBLAT3.SUMM
 test/cblat1
 test/cblat2
 test/cblat3
@@ -79,6 +80,7 @@ test/dblat3
 test/sblat1
 test/sblat2
 test/sblat3
+test/test_shgemm
 test/zblat1
 test/zblat2
 test/zblat3
diff --git a/.travis.yml b/.travis.yml
index 0f20aef5c..101147353 100644
--- a/.travis.yml
+++ b/.travis.yml
@@ -16,7 +16,6 @@ matrix:
       before_script: &common-before
         - COMMON_FLAGS="DYNAMIC_ARCH=1 TARGET=NEHALEM NUM_THREADS=32"
       script:
-        - set -e
         - make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
         - make -C test $COMMON_FLAGS $BTYPE
         - make -C ctest $COMMON_FLAGS $BTYPE
@@ -34,6 +33,16 @@ matrix:
         - TARGET_BOX=PPC64LE_LINUX
         - BTYPE="BINARY=64 USE_OPENMP=1"
 
+    - <<: *test-ubuntu
+      os: linux
+      arch: s390x
+      before_script:
+        - COMMON_FLAGS="DYNAMIC_ARCH=1 TARGET=Z13 NUM_THREADS=32"
+      env:
+        # for matrix annotation only
+        - TARGET_BOX=IBMZ_LINUX
+        - BTYPE="BINARY=64 USE_OPENMP=1"
+
     - <<: *test-ubuntu
       env:
         - TARGET_BOX=LINUX64
@@ -98,7 +107,6 @@ matrix:
         - sudo sh alpine-chroot-install -p 'build-base gfortran perl linux-headers'
       before_script: *common-before
       script:
-        - set -e
         # XXX: Disable some warnings for now to avoid exceeding Travis limit for log size.
         - alpine make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
               CFLAGS="-Wno-misleading-indentation -Wno-sign-conversion -Wno-incompatible-pointer-types"
@@ -141,7 +149,6 @@ matrix:
       before_script:
         - COMMON_ARGS="-DTARGET=NEHALEM -DNUM_THREADS=32"
       script:
-        - set -e
         - mkdir build
         - CONFIG=Release
         - cmake -Bbuild -H. $CMAKE_ARGS $COMMON_ARGS -DCMAKE_BUILD_TYPE=$CONFIG
@@ -180,6 +187,12 @@ matrix:
         - CFLAGS="-O2 -Wno-macro-redefined -isysroot /Applications/Xcode-10.1.app/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS12.1.sdk -arch arm64 -miphoneos-version-min=10.0"
         - BTYPE="TARGET=ARMV8 BINARY=64 HOSTCC=clang NOFORTRAN=1"
 
+    - <<: *test-macos
+      osx_image: xcode10.1
+      env:
+        - CC="/Applications/Xcode-10.1.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang"
+        - CFLAGS="-O2 -mno-thumb -Wno-macro-redefined -isysroot /Applications/Xcode-10.1.app/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS12.1.sdk -arch armv7 -miphoneos-version-min=5.1"
+        - BTYPE="TARGET=ARMV7 HOSTCC=clang NOFORTRAN=1"
 # whitelist
 branches:
   only:
diff --git a/CMakeLists.txt b/CMakeLists.txt
index eb1543591..12621d6b8 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -6,7 +6,8 @@ cmake_minimum_required(VERSION 2.8.5)
 project(OpenBLAS C ASM)
 set(OpenBLAS_MAJOR_VERSION 0)
 set(OpenBLAS_MINOR_VERSION 3)
-set(OpenBLAS_PATCH_VERSION 9)
+set(OpenBLAS_PATCH_VERSION 10)
+
 set(OpenBLAS_VERSION "${OpenBLAS_MAJOR_VERSION}.${OpenBLAS_MINOR_VERSION}.${OpenBLAS_PATCH_VERSION}")
 
 # Adhere to GNU filesystem layout conventions
@@ -23,6 +24,7 @@ option(BUILD_WITHOUT_CBLAS "Do not build the C interface (CBLAS) to the BLAS fun
 option(DYNAMIC_ARCH "Include support for multiple CPU targets, with automatic selection at runtime (x86/x86_64, aarch64 or ppc only)" OFF)
 option(DYNAMIC_OLDER "Include specific support for older x86 cpu models (Penryn,Dunnington,Atom,Nano,Opteron) with DYNAMIC_ARCH" OFF)
 option(BUILD_RELAPACK "Build with ReLAPACK (recursive implementation of several LAPACK functions on top of standard LAPACK)" OFF)
+option(USE_LOCKING "Use locks even in single-threaded builds to make them callable from multiple threads" OFF)
 if(${CMAKE_SYSTEM_NAME} MATCHES "Linux")
 option(NO_AFFINITY "Disable support for CPU affinity masks to avoid binding processes from e.g. R or numpy/scipy to a single core" ON)
 else()
@@ -86,9 +88,13 @@ if (NOT NO_LAPACK)
   list(APPEND SUBDIRS lapack)
 endif ()
 
+if (NOT DEFINED BUILD_HALF)
+ set (BUILD_HALF false)
+endif ()
 # set which float types we want to build for
 if (NOT DEFINED BUILD_SINGLE AND NOT DEFINED BUILD_DOUBLE AND NOT DEFINED BUILD_COMPLEX AND NOT DEFINED BUILD_COMPLEX16)
   # if none are defined, build for all
+#  set(BUILD_HALF true)
   set(BUILD_SINGLE true)
   set(BUILD_DOUBLE true)
   set(BUILD_COMPLEX true)
@@ -120,6 +126,11 @@ if (BUILD_COMPLEX16)
   list(APPEND FLOAT_TYPES "ZCOMPLEX") # defines COMPLEX and DOUBLE
 endif ()
 
+if (BUILD_HALF)
+  message(STATUS "Building Half Precision")
+  list(APPEND FLOAT_TYPES "HALF") # defines nothing
+endif ()
+
 if (NOT DEFINED CORE OR "${CORE}" STREQUAL "UNKNOWN")
   message(FATAL_ERROR "Detecting CPU failed. Please set TARGET explicitly, e.g. make TARGET=your_cpu_target. Please read README for details.")
 endif ()
@@ -234,7 +245,7 @@ if (BUILD_SHARED_LIBS AND BUILD_RELAPACK)
   if (NOT MSVC)
     target_link_libraries(${OpenBLAS_LIBNAME} "-Wl,-allow-multiple-definition")
   else()
-    target_link_libraries(${OpenBLAS_LIBNAME} "/FORCE:MULTIPLE")
+   set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} /FORCE:MULTIPLE")
   endif()
 endif()
 
diff --git a/CONTRIBUTORS.md b/CONTRIBUTORS.md
index df497c1d2..aba39e56f 100644
--- a/CONTRIBUTORS.md
+++ b/CONTRIBUTORS.md
@@ -180,3 +180,13 @@ In chronological order:
   * [2019-12-23] optimize AVX2 CGEMM and ZGEMM
   * [2019-12-30] AVX2 CGEMM3M & ZGEMM3M kernels
   * [2020-01-07] optimize AVX2 SGEMM and STRMM
+
+* Rajalakshmi Srinivasaraghavan <https://github.com/RajalakshmiSR>
+  * [2020-04-15] Half-precision GEMM for bfloat16
+
+* Marius Hillenbrand <https://github.com/mhillenibm>
+  * [2020-05-12] Revise dynamic architecture detection for IBM z
+  * [2020-05-12] Add new sgemm and strmm kernel for IBM z14
+
+* Danfeng Zhang <https://github.com/craft-zhang>
+  * [2020-05-20] Improve performance of SGEMM and STRMM on Arm Cortex-A53
\ No newline at end of file
diff --git a/Changelog.txt b/Changelog.txt
index 5f924629b..cbf0b50f5 100644
--- a/Changelog.txt
+++ b/Changelog.txt
@@ -1,4 +1,77 @@
 OpenBLAS ChangeLog
+====================================================================
+Version 0.3.10
+ 14-Jun-2020
+
+common:
+	* Improved thread locking behaviour in blas_server and parallel getrf                         
+	* Imported bugfix 394 from LAPACK (spurious reference to "XERBL"
+	  due to overlong lines)
+	* Imported bugfix 403 from LAPACK (compile option "recursive" required
+	  for correctness with Intel and PGI)
+	* Imported bugfix 408 from LAPACK (wrong scaling in ZHEEQUB)
+	* Imported bugfix 411 from LAPACK (infinite loop in LARGV/LARTG/LARTGP)
+	* Fixed mismatches between BUFFERSIZE and GEMM_UNROLL parameters that
+	  could lead to crashes at large matrix sizes  
+	* Restored internal soname in dynamic libraries on FreeBSD and Dragonfly 
+	* Added API (openblas_setaffinity) to set the thread affinity on Linux
+	* Added initial infrastructure for half-precision floating point 
+	  (bfloat16) support with a generic implementation of SHGEMM     
+	* Added CMAKE build system support for building the cblas_Xgemm3m
+	  functions
+	* Fixed CMAKE support for building in a path with embedded spaces
+	* Fixed CMAKE (non)handling of NO_EXPRECISION and MAX_STACK_ALLOC
+	* Fixed GCC version detection in the Makefiles
+	* Allowed overriding the names of AR, AS and LD in Makefile builds
+
+POWER:
+	* Fixed big-endian POWER8 ELFv2 builds on FreeBSD
+	* Fixed GCC version checks and DYNAMIC_ARCH builds on POWER9
+	* Fixed CMAKE build support for POWER9
+	* fixed a potential race condition in the thread buffer allocation
+	* Worked around LAPACK test failures on PPC G4
+
+MIPS:
+	* Fixed a potential race condition in the thread buffer allocation
+	* Added support for MIPS 24K/24KE family based on P5600 kernels
+
+MIPS64:
+	* fixed a potential race condition in the thread buffer allocation
+	* Added TARGET=GENERIC
+
+ARMV7:
+	* Fixed a race condition in the thread buffer allocation
+
+ARMV8:
+	* Fixed a race condition in the thread buffer allocation
+	* Fixed zero initialisation in the assembly for SGEMM and DGEMM BETA
+	* Improved performance of the ThunderX2 DAXPY kernel
+	* Added an optimized SGEMM kernel for Cortex A53    
+	* Fixed Makefile support for INTERFACE64 (8-byte integer)
+
+x86_64:
+	* Fixed a syntax error in the CMAKE setup for SkylakeX
+	* Improved performance of STRSM on Haswell, SkylakeX and Ryzen
+	* Improved SGEMM performance on SGEMM for workloads with ldc a
+	  multiple of 1024
+	* Improved DGEMM performance on Skylake X
+	* Fixed unwanted AVX512-dependency of SGEMM in DYNAMIC_ARCH
+	  builds created on SkylakeX
+	* Removed data alignment requirement in the SSE2 copy kernels
+	  that could cause spurious crashes
+	* Added a workaround for an optimizer bug in AppleClang 11.0.3
+	* Fixed LAPACK test failures due to wrong options for Intel Fortran
+	* Fixed compilation and LAPACK test results with recent Flang
+	  and AMD AOCC
+	* Fixed DYNAMIC_ARCH builds with CMAKE on OS X
+	* Fixed missing exports of cblas_i?amin, cblas_i?min, cblas_i?max,
+	  cblas_?sum, cblas_?gemm3m in the shared library on OS 
+	* Fixed reporting of cpu name in DYNAMIC_ARCH builds (would sometimes
+	  show the name of an older generation chip supported by the same kernels)
+
+IBM Z:
+	* Improved performance of SGEMM/STRMM and DGEMM/DTRMM on Z14
+
 ====================================================================
 Version 0.3.9
  1-Mar-2020
diff --git a/Makefile b/Makefile
index a22e16bab..e113026dd 100644
--- a/Makefile
+++ b/Makefile
@@ -56,10 +56,21 @@ ifneq ($(INTERFACE64), 0)
 	@echo "  Use 64 bits int    (equivalent to \"-i8\" in Fortran)      "
 endif
 endif
-
-	@echo "  C compiler       ... $(C_COMPILER)  (command line : $(CC))"
+	@$(CC) --version > /dev/null 2>&1;\
+	if [ $$? -eq 0 ]; then \
+	   cverinfo=`$(CC) --version | sed -n '1p'`; \
+	   echo "  C compiler       ... $(C_COMPILER)  (cmd & version : $${cverinfo})";\
+	else  \
+	   echo "  C compiler       ... $(C_COMPILER)  (command line : $(CC))";\
+	fi
 ifeq ($(NOFORTRAN), $(filter 0,$(NOFORTRAN)))
-	@echo "  Fortran compiler ... $(F_COMPILER)  (command line : $(FC))"
+	@$(FC) --version > /dev/null 2>&1;\
+	if [ $$? -eq 0 ]; then \
+	   fverinfo=`$(FC) --version | sed -n '1p'`; \
+	   echo "  Fortran compiler ... $(F_COMPILER)  (cmd & version : $${fverinfo})";\
+	else \
+	   echo "  Fortran compiler ... $(F_COMPILER)  (command line : $(FC))";\
+	fi
 endif
 ifneq ($(OSNAME), AIX)
 	@echo -n "  Library Name     ... $(LIBNAME)"
@@ -68,9 +79,13 @@ else
 endif
 
 ifndef SMP
-	@echo " (Single threaded)  "
+	@echo " (Single-threading)  "
 else
-	@echo " (Multi threaded; Max num-threads is $(NUM_THREADS))"
+	@echo " (Multi-threading; Max num-threads is $(NUM_THREADS))"
+endif
+
+ifeq ($(DYNAMIC_ARCH), 1)
+	@echo "  Supporting multiple $(ARCH) cpu models with minimum requirement for the common code being $(CORE)"
 endif
 
 ifeq ($(USE_OPENMP), 1)
@@ -97,12 +112,12 @@ endif
 
 shared :
 ifneq ($(NO_SHARED), 1)
-ifeq ($(OSNAME), $(filter $(OSNAME),Linux SunOS Android Haiku))
+ifeq ($(OSNAME), $(filter $(OSNAME),Linux SunOS Android Haiku FreeBSD DragonFly))
 	@$(MAKE) -C exports so
 	@ln -fs $(LIBSONAME) $(LIBPREFIX).so
 	@ln -fs $(LIBSONAME) $(LIBPREFIX).so.$(MAJOR_VERSION)
 endif
-ifeq ($(OSNAME), $(filter $(OSNAME),FreeBSD OpenBSD NetBSD DragonFly))
+ifeq ($(OSNAME), $(filter $(OSNAME),OpenBSD NetBSD))
 	@$(MAKE) -C exports so
 	@ln -fs $(LIBSONAME) $(LIBPREFIX).so
 endif
@@ -249,6 +264,7 @@ lapack_prebuild :
 ifeq ($(NOFORTRAN), $(filter 0,$(NOFORTRAN)))
 	-@echo "FC          = $(FC)" > $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "FFLAGS      = $(LAPACK_FFLAGS)" >> $(NETLIB_LAPACK_DIR)/make.inc
+	-@echo "FFLAGS_DRV  = $(LAPACK_FFLAGS)" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "POPTS       = $(LAPACK_FPFLAGS)" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "FFLAGS_NOOPT       = -O0 $(LAPACK_NOOPT)" >> $(NETLIB_LAPACK_DIR)/make.inc
 	-@echo "PNOOPT      = $(LAPACK_FPFLAGS) -O0" >> $(NETLIB_LAPACK_DIR)/make.inc
@@ -317,7 +333,7 @@ lapack-test :
 	$(MAKE) -j 1 -C $(NETLIB_LAPACK_DIR)/TESTING/EIG xeigtstc  xeigtstd  xeigtsts  xeigtstz 
 	$(MAKE) -j 1 -C $(NETLIB_LAPACK_DIR)/TESTING/LIN xlintstc  xlintstd  xlintstds  xlintstrfd  xlintstrfz  xlintsts  xlintstz  xlintstzc xlintstrfs xlintstrfc
 ifneq ($(CROSS), 1)
-	( cd $(NETLIB_LAPACK_DIR)/INSTALL; make all; ./testlsame; ./testslamch; ./testdlamch; \
+	( cd $(NETLIB_LAPACK_DIR)/INSTALL; $(MAKE) all; ./testlsame; ./testslamch; ./testdlamch; \
         ./testsecond; ./testdsecnd; ./testieee; ./testversion )
 	(cd $(NETLIB_LAPACK_DIR); ./lapack_testing.py -r -b TESTING)
 endif
diff --git a/Makefile.install b/Makefile.install
index 2dc32c3d9..dad869f4c 100644
--- a/Makefile.install
+++ b/Makefile.install
@@ -68,14 +68,14 @@ endif
 #for install shared library
 ifneq ($(NO_SHARED),1)
 	@echo Copying the shared library to $(DESTDIR)$(OPENBLAS_LIBRARY_DIR)
-ifeq ($(OSNAME), $(filter $(OSNAME),Linux SunOS Android Haiku))
+ifeq ($(OSNAME), $(filter $(OSNAME),Linux SunOS Android Haiku FreeBSD DragonFly))
 	@install -pm755 $(LIBSONAME) "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)"
 	@cd "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)" ; \
 	ln -fs $(LIBSONAME) $(LIBPREFIX).so ; \
 	ln -fs $(LIBSONAME) $(LIBPREFIX).so.$(MAJOR_VERSION)
 endif
 
-ifeq ($(OSNAME), $(filter $(OSNAME),FreeBSD OpenBSD NetBSD DragonFly))
+ifeq ($(OSNAME), $(filter $(OSNAME),OpenBSD NetBSD))
 	@cp $(LIBSONAME) "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)"
 	@cd "$(DESTDIR)$(OPENBLAS_LIBRARY_DIR)" ; \
 	ln -fs $(LIBSONAME) $(LIBPREFIX).so
diff --git a/Makefile.prebuild b/Makefile.prebuild
index b00f13368..48fb5e991 100644
--- a/Makefile.prebuild
+++ b/Makefile.prebuild
@@ -17,7 +17,11 @@ ifdef CPUIDEMU
 EXFLAGS = -DCPUIDEMU -DVENDOR=99
 endif
 
-ifeq ($(TARGET), 1004K)
+ifeq ($(TARGET), MIPS24K)
+TARGET_FLAGS = -mips32r2
+endif
+
+ifeq ($(TARGET), MIPS1004K)
 TARGET_FLAGS = -mips32r2
 endif
 
diff --git a/Makefile.rule b/Makefile.rule
index a4465e448..551c094ca 100644
--- a/Makefile.rule
+++ b/Makefile.rule
@@ -3,7 +3,7 @@
 #
 
 # This library's version
-VERSION = 0.3.9
+VERSION = 0.3.10
 
 # If you set the suffix, the library name will be libopenblas_$(LIBNAMESUFFIX).a
 # and libopenblas_$(LIBNAMESUFFIX).so. Meanwhile, the soname in shared library
@@ -273,6 +273,9 @@ COMMON_PROF = -pg
 #
 # CPP_THREAD_SAFETY_TEST = 1
 
+
+# If you want to enable the experimental BFLOAT16 support
+# BUILD_HALF = 1
 #
 #  End of user configuration
 #
diff --git a/Makefile.system b/Makefile.system
index 1e30d05a8..8d78b420f 100644
--- a/Makefile.system
+++ b/Makefile.system
@@ -21,6 +21,8 @@ ifeq ($(ARCH), amd64)
 override ARCH=x86_64
 else ifeq ($(ARCH), powerpc64)
 override ARCH=power
+else ifeq ($(ARCH), powerpc)
+override ARCH=power
 else ifeq ($(ARCH), i386)
 override ARCH=x86
 else ifeq ($(ARCH), aarch64)
@@ -261,10 +263,10 @@ endif
 
 ARFLAGS	=
 CPP	= $(COMPILER) -E
-AR	= $(CROSS_SUFFIX)ar
-AS	= $(CROSS_SUFFIX)as
-LD	= $(CROSS_SUFFIX)ld
-RANLIB	= $(CROSS_SUFFIX)ranlib
+AR	?= $(CROSS_SUFFIX)ar
+AS	?= $(CROSS_SUFFIX)as
+LD	?= $(CROSS_SUFFIX)ld
+RANLIB	?= $(CROSS_SUFFIX)ranlib
 NM	= $(CROSS_SUFFIX)nm
 DLLWRAP = $(CROSS_SUFFIX)dllwrap
 OBJCOPY = $(CROSS_SUFFIX)objcopy
@@ -277,6 +279,15 @@ NO_LAPACK = 1
 override FEXTRALIB = 
 endif
 
+ifeq ($(C_COMPILER), GCC)
+GCCVERSIONGTEQ4 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 4)
+GCCVERSIONGT4 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \> 4)
+GCCVERSIONGT5 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \> 5)
+GCCVERSIONGTEQ7 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 7)
+GCCVERSIONGTEQ9 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 9)
+GCCMINORVERSIONGTEQ7 := $(shell expr `$(CC) -dumpversion | cut -f2 -d.` \>= 7)
+endif
+
 #
 #  OS dependent settings
 #
@@ -323,14 +334,7 @@ ifeq ($(C_COMPILER), CLANG)
 CCOMMON_OPT	+= -DMS_ABI
 endif
 
-ifeq ($(C_COMPILER), GCC)
 #Version tests for supporting specific features (MS_ABI, POWER9 intrinsics)
-GCCVERSIONGTEQ4 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 4)
-GCCVERSIONGT4 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \> 4)
-GCCVERSIONGT5 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \> 5)
-GCCVERSIONGTEQ7 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 7)
-GCCVERSIONGTEQ9 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 9)
-GCCMINORVERSIONGTEQ7 := $(shell expr `$(CC) -dumpversion | cut -f2 -d.` \>= 7)
 ifeq ($(GCCVERSIONGT4), 1)
 # GCC Major version > 4
 # It is compatible with MSVC ABI.
@@ -344,7 +348,6 @@ ifeq ($(GCCMINORVERSIONGTEQ7), 1)
 CCOMMON_OPT	+= -DMS_ABI
 endif
 endif
-endif
 
 # Ensure the correct stack alignment on Win32
 # http://permalink.gmane.org/gmane.comp.lib.openblas.general/97
@@ -564,8 +567,27 @@ DYNAMIC_CORE += EMAG8180
 endif
 
 ifeq ($(ARCH), zarch)
-DYNAMIC_CORE = Z13
+DYNAMIC_CORE = ZARCH_GENERIC
+
+# Z13 is supported since gcc-5.2, gcc-6, and in RHEL 7.3 and newer
+GCC_GE_52 := $(subst 0,,$(shell expr `$(CC) -dumpversion` \>= "5.2"))
+
+ifeq ($(wildcard /etc/redhat-release), /etc/redhat-release)
+RHEL_WITH_Z13 := $(subst 0,,$(shell source /etc/os-release ; expr $$VERSION_ID \>= "7.3"))
+endif
+
+ifeq ($(or $(GCC_GE_52),$(RHEL_WITH_Z13)), 1)
+DYNAMIC_CORE += Z13
+else
+$(info OpenBLAS: Not building Z13 kernels because gcc is older than 5.2 or 6.x)
+endif
+
+GCC_MAJOR_GE_7 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 7)
+ifeq ($(GCC_MAJOR_GE_7), 1)
 DYNAMIC_CORE += Z14
+else
+$(info OpenBLAS: Not building Z14 kernels because gcc is older than 7.x)
+endif
 endif
 
 ifeq ($(ARCH), power)
@@ -575,6 +597,7 @@ ifneq ($(C_COMPILER), GCC)
 DYNAMIC_CORE += POWER9
 endif
 ifeq ($(C_COMPILER), GCC)
+GCCVERSIONGT5 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \> 5)
 ifeq ($(GCCVERSIONGT5), 1)
 DYNAMIC_CORE += POWER9
 else
@@ -635,6 +658,16 @@ endif
 ifeq ($(ARCH), arm64)
 NO_BINARY_MODE  = 1
 BINARY_DEFINED  = 1
+ifdef INTERFACE64
+ifneq ($(INTERFACE64), 0)
+ifeq ($(F_COMPILER), GFORTRAN)
+FCOMMON_OPT +=  -fdefault-integer-8
+endif
+ifeq ($(F_COMPILER), FLANG)
+FCOMMON_OPT += -i8
+endif
+endif
+endif
 endif
 
 
@@ -680,7 +713,12 @@ CCOMMON_OPT += -march=mips64
 FCOMMON_OPT += -march=mips64
 endif
 
-ifeq ($(CORE), 1004K)
+ifeq ($(CORE), MIPS24K)
+CCOMMON_OPT += -mips32r2 -mtune=24kc  $(MSA_FLAGS)
+FCOMMON_OPT += -mips32r2 -mtune=24kc  $(MSA_FLAGS)
+endif
+
+ifeq ($(CORE), MIPS1004K)
 CCOMMON_OPT += -mips32r2   $(MSA_FLAGS)
 FCOMMON_OPT += -mips32r2   $(MSA_FLAGS)
 endif
@@ -745,6 +783,15 @@ endif
 
 ifeq ($(F_COMPILER), FLANG)
 CCOMMON_OPT += -DF_INTERFACE_FLANG
+FCOMMON_OPT += -Mrecursive -Kieee
+ifeq ($(OSNAME), Linux)
+ifeq ($(ARCH), x86_64)
+FLANG_VENDOR := $(shell expr `$(FC) --version|cut -f 1 -d "."|head -1`)
+ifeq ($(FLANG_VENDOR),AOCC)
+FCOMMON_OPT += -fno-unroll-loops
+endif
+endif
+endif
 ifdef BINARY64
 ifdef INTERFACE64
 ifneq ($(INTERFACE64), 0)
@@ -840,6 +887,7 @@ ifneq ($(INTERFACE64), 0)
 FCOMMON_OPT += -i8
 endif
 endif
+FCOMMON_OPT += -recursive -fp-model strict -assume protect-parens
 ifeq ($(USE_OPENMP), 1)
 FCOMMON_OPT += -fopenmp
 endif
@@ -883,6 +931,7 @@ FCOMMON_OPT += -tp p7-64
 else
 FCOMMON_OPT += -tp p7
 endif
+FCOMMON_OPT += -Mrecursive
 ifeq ($(USE_OPENMP), 1)
 FCOMMON_OPT += -mp
 endif
@@ -1107,6 +1156,10 @@ ifeq ($(USE_TLS), 1)
 CCOMMON_OPT += -DUSE_TLS
 endif
 
+ifeq ($(BUILD_HALF), 1)
+CCOMMON_OPT += -DBUILD_HALF
+endif
+
 CCOMMON_OPT += -DVERSION=\"$(VERSION)\"
 
 ifndef SYMBOLPREFIX
@@ -1133,6 +1186,7 @@ KERNELDIR	= $(TOPDIR)/kernel/$(ARCH)
 
 include $(TOPDIR)/Makefile.$(ARCH)
 
+CCOMMON_OPT     += -UASMNAME -UASMFNAME -UNAME -UCNAME -UCHAR_NAME -UCHAR_CNAME
 CCOMMON_OPT	+= -DASMNAME=$(FU)$(*F) -DASMFNAME=$(FU)$(*F)$(BU) -DNAME=$(*F)$(BU) -DCNAME=$(*F) -DCHAR_NAME=\"$(*F)$(BU)\" -DCHAR_CNAME=\"$(*F)\"
 
 ifeq ($(CORE), PPC440)
@@ -1225,7 +1279,6 @@ endif
 
 override CFLAGS     += $(COMMON_OPT) $(CCOMMON_OPT) -I$(TOPDIR)
 override PFLAGS     += $(COMMON_OPT) $(CCOMMON_OPT) -I$(TOPDIR) -DPROFILE $(COMMON_PROF)
-
 override FFLAGS     += $(COMMON_OPT) $(FCOMMON_OPT)
 override FPFLAGS    += $(FCOMMON_OPT) $(COMMON_PROF)
 #MAKEOVERRIDES =
@@ -1332,6 +1385,7 @@ export ARCH
 export CORE
 export LIBCORE
 export __BYTE_ORDER__
+export ELF_VERSION
 export PGCPATH
 export CONFIG
 export CC
@@ -1377,7 +1431,10 @@ export KERNELDIR
 export FUNCTION_PROFILE
 export TARGET_CORE
 export NO_AVX512
+export BUILD_HALF
 
+export SHGEMM_UNROLL_M
+export SHGEMM_UNROLL_N
 export SGEMM_UNROLL_M
 export SGEMM_UNROLL_N
 export DGEMM_UNROLL_M
diff --git a/Makefile.tail b/Makefile.tail
index 2adede1a5..39902982b 100644
--- a/Makefile.tail
+++ b/Makefile.tail
@@ -1,3 +1,4 @@
+SHBLASOBJS_P = $(SHBLASOBJS:.$(SUFFIX)=.$(PSUFFIX))
 SBLASOBJS_P = $(SBLASOBJS:.$(SUFFIX)=.$(PSUFFIX))
 DBLASOBJS_P = $(DBLASOBJS:.$(SUFFIX)=.$(PSUFFIX))
 QBLASOBJS_P = $(QBLASOBJS:.$(SUFFIX)=.$(PSUFFIX))
@@ -9,8 +10,8 @@ COMMONOBJS_P = $(COMMONOBJS:.$(SUFFIX)=.$(PSUFFIX))
 
 HPLOBJS_P   = $(HPLOBJS:.$(SUFFIX)=.$(PSUFFIX))
 
-BLASOBJS    = $(SBLASOBJS)   $(DBLASOBJS)   $(CBLASOBJS)   $(ZBLASOBJS)
-BLASOBJS_P  = $(SBLASOBJS_P) $(DBLASOBJS_P) $(CBLASOBJS_P) $(ZBLASOBJS_P)
+BLASOBJS    = $(SHBLASOBJS)  $(SBLASOBJS)   $(DBLASOBJS)   $(CBLASOBJS)   $(ZBLASOBJS)
+BLASOBJS_P  = $(SHBLASOBJS_P) $(SBLASOBJS_P) $(DBLASOBJS_P) $(CBLASOBJS_P) $(ZBLASOBJS_P)
 
 ifdef EXPRECISION
 BLASOBJS   += $(QBLASOBJS)   $(XBLASOBJS)
@@ -22,6 +23,7 @@ BLASOBJS   += $(QBLASOBJS)   $(XBLASOBJS)
 BLASOBJS_P += $(QBLASOBJS_P) $(XBLASOBJS_P)
 endif
 
+$(SHBLASOBJS) $(SHBLASOBJS_P) : override CFLAGS += -DHALF -UDOUBLE  -UCOMPLEX
 $(SBLASOBJS) $(SBLASOBJS_P) : override CFLAGS += -UDOUBLE  -UCOMPLEX
 $(DBLASOBJS) $(DBLASOBJS_P) : override CFLAGS += -DDOUBLE  -UCOMPLEX
 $(QBLASOBJS) $(QBLASOBJS_P) : override CFLAGS += -DXDOUBLE -UCOMPLEX
@@ -29,6 +31,7 @@ $(CBLASOBJS) $(CBLASOBJS_P) : override CFLAGS += -UDOUBLE  -DCOMPLEX
 $(ZBLASOBJS) $(ZBLASOBJS_P) : override CFLAGS += -DDOUBLE  -DCOMPLEX
 $(XBLASOBJS) $(XBLASOBJS_P) : override CFLAGS += -DXDOUBLE -DCOMPLEX
 
+$(SHBLASOBJS_P) : override CFLAGS += -DPROFILE $(COMMON_PROF)
 $(SBLASOBJS_P) : override CFLAGS += -DPROFILE $(COMMON_PROF)
 $(DBLASOBJS_P) : override CFLAGS += -DPROFILE $(COMMON_PROF)
 $(QBLASOBJS_P) : override CFLAGS += -DPROFILE $(COMMON_PROF)
diff --git a/Makefile.zarch b/Makefile.zarch
index 47ea1eb71..be1e34f6d 100644
--- a/Makefile.zarch
+++ b/Makefile.zarch
@@ -5,6 +5,6 @@ FCOMMON_OPT += -march=z13 -mzvector
 endif
 
 ifeq ($(CORE), Z14)
-CCOMMON_OPT += -march=z14 -mzvector
+CCOMMON_OPT += -march=z14 -mzvector -O3
 FCOMMON_OPT += -march=z14 -mzvector
 endif
diff --git a/README.md b/README.md
index 04f43f4c7..6dc3c7b42 100644
--- a/README.md
+++ b/README.md
@@ -6,8 +6,11 @@ Travis CI: [![Build Status](https://travis-ci.org/xianyi/OpenBLAS.svg?branch=dev
 
 AppVeyor: [![Build status](https://ci.appveyor.com/api/projects/status/09sohd35n8nkkx64/branch/develop?svg=true)](https://ci.appveyor.com/project/xianyi/openblas/branch/develop)
 
+Drone CI: [![Build Status](https://cloud.drone.io/api/badges/xianyi/OpenBLAS/status.svg?branch=develop)](https://cloud.drone.io/xianyi/OpenBLAS/)
+
 [![Build Status](https://dev.azure.com/xianyi/OpenBLAS/_apis/build/status/xianyi.OpenBLAS?branchName=develop)](https://dev.azure.com/xianyi/OpenBLAS/_build/latest?definitionId=1&branchName=develop)
 
+
 ## Introduction
 
 OpenBLAS is an optimized BLAS library based on GotoBLAS2 1.13 BSD version.
@@ -119,6 +122,11 @@ Please read `GotoBLAS_01Readme.txt` for older CPU models already supported by th
 - **AMD STEAMROLLER**: Uses Bulldozer codes with some optimizations.
 - **AMD ZEN**: Uses Haswell codes with some optimizations.
 
+#### MIPS32
+
+- **MIPS 1004K**: uses P5600 codes
+- **MIPS 24K**: uses P5600 codes
+
 #### MIPS64
 
 - **ICT Loongson 3A**: Optimized Level-3 BLAS and the part of Level-1,2.
@@ -140,6 +148,7 @@ Please read `GotoBLAS_01Readme.txt` for older CPU models already supported by th
 - **ThunderX**: Optimized some Level-1 functions
 - **ThunderX2T99**: Optimized Level-3 BLAS and parts of Levels 1 and 2
 - **TSV110**: Optimized some Level-3 helper functions
+- **EMAG 8180**: preliminary support based on A57
 
 #### PPC/PPC64
 
@@ -154,11 +163,16 @@ Please read `GotoBLAS_01Readme.txt` for older CPU models already supported by th
 ### Support for multiple targets in a single library
 
 OpenBLAS can be built for multiple targets with runtime detection of the target cpu by specifiying DYNAMIC_ARCH=1 in Makefile.rule, on the gmake command line or as -DDYNAMIC_ARCH=TRUE in cmake.
+
 For **x86_64**, the list of targets this activates contains Prescott, Core2, Nehalem, Barcelona, Sandybridge, Bulldozer, Piledriver, Steamroller, Excavator, Haswell, Zen, SkylakeX. For cpu generations not included in this list, the corresponding older model is used. If you also specify DYNAMIC_OLDER=1, specific support for Penryn, Dunnington, Opteron, Opteron/SSE3, Bobcat, Atom and Nano is added. Finally there is an option DYNAMIC_LIST that allows to specify an individual list of targets to include instead of the default.
+
 DYNAMIC_ARCH is also supported on **x86**, where it translates to Katmai, Coppermine, Northwood, Prescott, Banias,
 Core2, Penryn, Dunnington, Nehalem, Athlon, Opteron, Opteron_SSE3, Barcelona, Bobcat, Atom and Nano.
+
 On **ARMV8**, it enables support for CortexA53, CortexA57, CortexA72, CortexA73, Falkor, ThunderX, ThunderX2T99, TSV110 as well as generic ARMV8 cpus.
+
 For **POWER**, the list encompasses POWER6, POWER8 and POWER9, on **ZARCH** it comprises Z13 and Z14.
+
 The TARGET option can be used in conjunction with DYNAMIC_ARCH=1 to specify which cpu model should be assumed for all the
 common code in the library, usually you will want to set this to the oldest model you expect to encounter.
 Please note that it is not possible to combine support for different architectures, so no combined 32 and 64 bit or x86_64 and arm64 in the same library.
diff --git a/TargetList.txt b/TargetList.txt
index f4a40ed02..e2d2f4026 100644
--- a/TargetList.txt
+++ b/TargetList.txt
@@ -58,7 +58,8 @@ CELL
 
 3.MIPS CPU:
 P5600
-1004K
+MIPS1004K
+MIPS24K
 
 4.MIPS64 CPU:
 SICORTEX
diff --git a/azure-pipelines.yml b/azure-pipelines.yml
index 9b4c85367..639cb3558 100644
--- a/azure-pipelines.yml
+++ b/azure-pipelines.yml
@@ -49,3 +49,23 @@ jobs:
       # we need a privileged docker run for sde process attachment
       docker run --privileged intel_sde
     displayName: 'Run AVX512 SkylakeX docker build / test'
+
+- job: Windows_cl
+  pool:
+     vmImage: 'windows-latest'
+  steps:   
+  - task: CMake@1
+    inputs:
+      workingDirectory: 'build' # Optional
+      cmakeArgs: '-G "Visual Studio 16 2019" ..'
+  - task: CMake@1
+    inputs:
+      cmakeArgs: '--build . --config Release'
+      workingDirectory: 'build'
+  - script: |
+      cd build
+      cd utest
+      dir
+      openblas_utest.exe
+  
+      
diff --git a/benchmark/Makefile b/benchmark/Makefile
index 1d4a220e4..2f70ceaf3 100644
--- a/benchmark/Makefile
+++ b/benchmark/Makefile
@@ -1,2392 +1,3439 @@
-TOPDIR	= ..
-include $(TOPDIR)/Makefile.system
-
-# ACML standard
-#ACML=/opt/acml5.3.1/gfortran64_mp/lib
-#LIBACML = -fopenmp $(ACML)/libacml_mp.a -lgfortran -lm
-
-# ACML custom
-#ACML=/opt/pb/acml-5-3-1-gfortran-64bit/gfortran64_fma4_mp/lib
-#LIBACML = -fopenmp $(ACML)/libacml_mp.a -lgfortran -lm
-
-# ACML 6.1 custom
-ACML=/home/saar/acml6.1/gfortran64_mp/lib
-LIBACML = -fopenmp $(ACML)/libacml_mp.so -lgfortran -lm
-
-
-# Atlas Ubuntu
-#ATLAS=/usr/lib/atlas-base
-#LIBATLAS = -fopenmp $(ATLAS)/liblapack_atlas.a  $(ATLAS)/libptcblas.a  $(ATLAS)/libptf77blas.a  $(ATLAS)/libatlas.a -lgfortran -lm
-
-# Atlas RHEL and Fedora
-ATLAS=/usr/lib64/atlas
-LIBATLAS = -fopenmp $(ATLAS)/liblapack.a  $(ATLAS)/libptcblas.a  $(ATLAS)/libptf77blas.a  $(ATLAS)/libatlas.a -lgfortran -lm
-
-# Intel standard
-# MKL=/opt/intel/mkl/lib/intel64
-# LIBMKL = -L$(MKL) -lmkl_intel_lp64 -lmkl_gnu_thread -lmkl_core -lgomp -lpthread -lm
-
-# Intel custom
-MKL=/home/saar/intel_mkl
-LIBMKL = -L$(MKL) -lmkl_intel_lp64 -lmkl_gnu_thread -lmkl_core -lgomp -lpthread -lm
-
-# Apple vecLib
-LIBVECLIB = -framework Accelerate
-
-ESSL=/opt/ibm/lib
-#LIBESSL = -lesslsmp $(ESSL)/libxlomp_ser.so.1 $(ESSL)/libxlf90_r.so.1 $(ESSL)/libxlfmath.so.1 $(ESSL)/libxlsmp.so.1 /opt/ibm/xlC/13.1.3/lib/libxl.a
-LIBESSL = -lesslsmp  $(ESSL)/libxlf90_r.so.1 $(ESSL)/libxlfmath.so.1 $(ESSL)/libxlsmp.so.1 /opt/ibm/xlC/13.1.3/lib/libxl.a
-
-ifneq ($(NO_LAPACK), 1)
-GOTO_LAPACK_TARGETS=slinpack.goto dlinpack.goto clinpack.goto zlinpack.goto \
-		    scholesky.goto dcholesky.goto ccholesky.goto zcholesky.goto \
-		    sgesv.goto dgesv.goto cgesv.goto zgesv.goto \
-		    sgeev.goto dgeev.goto cgeev.goto zgeev.goto \
-		    csymv.goto zsymv.goto \
-		    sgetri.goto dgetri.goto cgetri.goto zgetri.goto \
-		    spotrf.goto dpotrf.goto cpotrf.goto zpotrf.goto
-else
-GOTO_LAPACK_TARGETS=
-endif
-
-ifeq ($(OSNAME), WINNT)
-
-goto :: slinpack.goto dlinpack.goto clinpack.goto zlinpack.goto \
-       scholesky.goto dcholesky.goto ccholesky.goto zcholesky.goto \
-       sgemm.goto dgemm.goto cgemm.goto zgemm.goto \
-       strmm.goto dtrmm.goto ctrmm.goto ztrmm.goto \
-       strsm.goto dtrsm.goto ctrsm.goto ztrsm.goto \
-       ssyr.goto dsyr.goto \
-       ssyrk.goto dsyrk.goto csyrk.goto zsyrk.goto \
-       ssyr2k.goto dsyr2k.goto csyr2k.goto zsyr2k.goto \
-       sger.goto dger.goto cger.goto zger.goto \
-       sdot.goto ddot.goto \
-       srot.goto drot.goto \
-       saxpy.goto daxpy.goto caxpy.goto zaxpy.goto \
-       scopy.goto dcopy.goto ccopy.goto zcopy.goto \
-       sswap.goto dswap.goto cswap.goto zswap.goto \
-       sscal.goto dscal.goto cscal.goto zscal.goto \
-       sasum.goto dasum.goto casum.goto zasum.goto \
-       ssymv.goto dsymv.goto csymv.goto zsymv.goto \
-       chemv.goto zhemv.goto \
-       chemm.goto zhemm.goto \
-       cherk.goto zherk.goto \
-       cher2k.goto zher2k.goto \
-       sgemv.goto dgemv.goto cgemv.goto zgemv.goto \
-       sgeev.goto dgeev.goto cgeev.goto zgeev.goto \
-       sgesv.goto dgesv.goto cgesv.goto zgesv.goto \
-       sgetri.goto dgetri.goto cgetri.goto zgetri.goto \
-       spotrf.goto dpotrf.goto cpotrf.goto zpotrf.goto \
-       ssymm.goto dsymm.goto csymm.goto zsymm.goto
-
-acml :: slinpack.acml dlinpack.acml clinpack.acml zlinpack.acml \
-       scholesky.acml dcholesky.acml ccholesky.acml zcholesky.acml \
-       sgemm.acml dgemm.acml cgemm.acml zgemm.acml \
-       strmm.acml dtrmm.acml ctrmm.acml ztrmm.acml \
-       strsm.acml dtrsm.acml ctrsm.acml ztrsm.acml \
-       ssyr.acml dsyr.acml \
-       ssyrk.acml dsyrk.acml csyrk.acml zsyrk.acml \
-       ssyr2k.acml dsyr2k.acml csyr2k.acml zsyr2k.acml \
-       sger.acml dger.acml cger.acml zger.acml \
-       sdot.acml ddot.acml \
-       saxpy.acml daxpy.acml caxpy.acml zaxpy.acml \
-       scopy.acml dcopy.acml ccopy.acml zcopy.acml \
-       sswap.acml dswap.acml cswap.acml zswap.acml \
-       sscal.acml dscal.acml cscal.acml zscal.acml \
-       sasum.acml dasum.acml casum.acml zasum.acml \
-       ssymv.acml dsymv.acml csymv.acml zsymv.acml \
-       chemv.acml zhemv.acml \
-       chemm.acml zhemm.acml \
-       cherk.acml zherk.acml \
-       cher2k.acml zher2k.acml \
-       sgemv.acml dgemv.acml cgemv.acml zgemv.acml \
-       sgeev.acml dgeev.acml cgeev.acml zgeev.acml \
-       sgesv.acml dgesv.acml cgesv.acml zgesv.acml \
-       sgetri.acml dgetri.acml cgetri.acml zgetri.acml \
-       spotrf.acml dpotrf.acml cpotrf.acml zpotrf.acml \
-       ssymm.acml dsymm.acml csymm.acml zsymm.acml
-
-atlas :: slinpack.atlas dlinpack.atlas clinpack.atlas zlinpack.atlas \
-       scholesky.atlas dcholesky.atlas ccholesky.atlas zcholesky.atlas \
-       sgemm.atlas dgemm.atlas cgemm.atlas zgemm.atlas \
-       strmm.atlas dtrmm.atlas ctrmm.atlas ztrmm.atlas \
-       strsm.atlas dtrsm.atlas ctrsm.atlas ztrsm.atlas \
-       ssyr.goto dsyr.atlas \
-       ssyrk.atlas dsyrk.atlas csyrk.atlas zsyrk.atlas \
-       ssyr2k.atlas dsyr2k.atlas csyr2k.atlas zsyr2k.atlas \
-       sger.atlas dger.atlas cger.atlas zger.atlas\
-       sdot.atlas ddot.atlas \
-       saxpy.atlas daxpy.atlas caxpy.atlas zaxpy.atlas \
-       scopy.atlas dcopy.atlas ccopy.atlas zcopy.atlas \
-       sswap.atlas dswap.atlas cswap.atlas zswap.atlas \
-       sscal.atlas dscal.atlas cscal.atlas zscal.atlas \
-       sasum.atlas dasum.atlas casum.atlas zasum.atlas \
-       ssymv.atlas dsymv.atlas csymv.atlas zsymv.atlas \
-       chemv.atlas zhemv.atlas \
-       chemm.acml zhemm.acml \
-       chemm.atlas zhemm.atlas \
-       cherk.atlas zherk.atlas \
-       cher2k.atlas zher2k.atlas \
-       sgemv.atlas dgemv.atlas cgemv.atlas zgemv.atlas \
-       sgeev.atlas dgeev.atlas cgeev.atlas zgeev.atlas \
-       sgesv.atlas dgesv.atlas cgesv.atlas zgesv.atlas \
-       sgetri.atlas dgetri.atlas cgetri.atlas zgetri.atlas \
-       spotrf.atlas dpotrf.atlas cpotrf.atlas zpotrf.atlas \
-       ssymm.atlas dsymm.atlas csymm.atlas zsymm.atlas
-
-mkl :: slinpack.mkl dlinpack.mkl clinpack.mkl zlinpack.mkl \
-       scholesky.mkl dcholesky.mkl ccholesky.mkl zcholesky.mkl \
-       sgemm.mkl dgemm.mkl cgemm.mkl zgemm.mkl \
-       strmm.mkl dtrmm.mkl ctrmm.mkl ztrmm.mkl \
-       strsm.mkl dtrsm.mkl ctrsm.mkl ztrsm.mkl \
-       ssyr.mkl dsyr.mkl \
-       ssyrk.mkl dsyrk.mkl csyrk.mkl zsyrk.mkl \
-       ssyr2k.mkl dsyr2k.mkl csyr2k.mkl zsyr2k.mkl \
-       sger.mkl dger.mkl cger.mkl zger.mkl \
-       sdot.mkl ddot.mkl \
-       saxpy.mkl daxpy.mkl caxpy.mkl zaxpy.mkl \
-       scopy.mkl dcopy.mkl ccopy.mkl zcopy.mkl \
-       sswap.mkl dswap.mkl cswap.mkl zswap.mkl \
-       sscal.mkl dscal.mkl cscal.mkl zscal.mkl \
-       sasum.mkl dasum.mkl casum.mkl zasum.mkl \
-       ssymv.mkl dsymv.mkl csymv.mkl zsymv.mkl \
-       chemv.mkl zhemv.mkl \
-       chemm.mkl zhemm.mkl \
-       cherk.mkl zherk.mkl \
-       cher2k.mkl zher2k.mkl \
-       sgemv.mkl dgemv.mkl cgemv.mkl zgemv.mkl \
-       sgeev.mkl dgeev.mkl cgeev.mkl zgeev.mkl \
-       sgesv.mkl dgesv.mkl cgesv.mkl zgesv.mkl \
-       sgetri.mkl dgetri.mkl cgetri.mkl zgetri.mkl \
-       spotrf.mkl dpotrf.mkl cpotrf.mkl zpotrf.mkl \
-       ssymm.mkl dsymm.mkl csymm.mkl zsymm.mkl
-
-else
-
-goto :: sgemm.goto dgemm.goto cgemm.goto zgemm.goto \
-       strmm.goto dtrmm.goto ctrmm.goto ztrmm.goto \
-       strsm.goto dtrsm.goto ctrsm.goto ztrsm.goto \
-       ssyr.goto dsyr.goto \
-       ssyrk.goto dsyrk.goto csyrk.goto zsyrk.goto \
-       ssyr2k.goto dsyr2k.goto csyr2k.goto zsyr2k.goto \
-       sger.goto dger.goto cger.goto zger.goto \
-       sdot.goto ddot.goto cdot.goto zdot.goto \
-       srot.goto drot.goto \
-       saxpy.goto daxpy.goto caxpy.goto zaxpy.goto \
-       scopy.goto dcopy.goto ccopy.goto zcopy.goto \
-       sswap.goto dswap.goto cswap.goto zswap.goto \
-       sscal.goto dscal.goto cscal.goto zscal.goto \
-       sasum.goto dasum.goto casum.goto zasum.goto \
-       ssymv.goto dsymv.goto \
-       chemv.goto zhemv.goto \
-       chemm.goto zhemm.goto \
-       cherk.goto zherk.goto \
-       cher2k.goto zher2k.goto \
-       sgemv.goto dgemv.goto cgemv.goto zgemv.goto \
-       ssymm.goto dsymm.goto csymm.goto zsymm.goto \
-       smallscaling \
-       isamax.goto idamax.goto icamax.goto izamax.goto \
-       snrm2.goto dnrm2.goto scnrm2.goto dznrm2.goto $(GOTO_LAPACK_TARGETS)
-
-acml :: slinpack.acml dlinpack.acml clinpack.acml zlinpack.acml \
-       scholesky.acml dcholesky.acml ccholesky.acml zcholesky.acml \
-       sgemm.acml dgemm.acml cgemm.acml zgemm.acml \
-       strmm.acml dtrmm.acml ctrmm.acml ztrmm.acml \
-       strsm.acml dtrsm.acml ctrsm.acml ztrsm.acml \
-       ssyr.acml dsyr.acml \
-       ssyrk.acml dsyrk.acml csyrk.acml zsyrk.acml \
-       ssyr2k.acml dsyr2k.acml csyr2k.acml zsyr2k.acml \
-       sger.acml dger.acml cger.acml zger.acml \
-       sdot.acml ddot.acml \
-       saxpy.acml daxpy.acml caxpy.acml zaxpy.acml \
-       scopy.acml dcopy.acml ccopy.acml zcopy.acml \
-       sswap.acml dswap.acml cswap.acml zswap.acml \
-       sscal.acml dscal.acml cscal.acml zscal.acml \
-       sasum.acml dasum.acml casum.acml zasum.acml \
-       ssymv.acml dsymv.acml csymv.acml zsymv.acml \
-       chemv.acml zhemv.acml \
-       chemm.acml zhemm.acml \
-       cherk.acml zherk.acml \
-       cher2k.acml zher2k.acml \
-       sgemv.acml dgemv.acml cgemv.acml zgemv.acml \
-       sgeev.acml dgeev.acml cgeev.acml zgeev.acml \
-       sgesv.acml dgesv.acml cgesv.acml zgesv.acml \
-       sgetri.acml dgetri.acml cgetri.acml zgetri.acml \
-       spotrf.acml dpotrf.acml cpotrf.acml zpotrf.acml \
-       ssymm.acml dsymm.acml csymm.acml zsymm.acml
-
-atlas :: slinpack.atlas dlinpack.atlas clinpack.atlas zlinpack.atlas \
-       scholesky.atlas dcholesky.atlas ccholesky.atlas zcholesky.atlas \
-       sgemm.atlas dgemm.atlas cgemm.atlas zgemm.atlas \
-       strmm.atlas dtrmm.atlas ctrmm.atlas ztrmm.atlas \
-       strsm.atlas dtrsm.atlas ctrsm.atlas ztrsm.atlas \
-       ssyr.atlas dsyr.atlas \
-       ssyrk.atlas dsyrk.atlas csyrk.atlas zsyrk.atlas \
-       ssyr2k.atlas dsyr2k.atlas csyr2k.atlas zsyr2k.atlas \
-       sger.atlas dger.atlas cger.atlas zger.atlas\
-       sdot.atlas ddot.atlas \
-       saxpy.atlas daxpy.atlas caxpy.atlas zaxpy.atlas \
-       scopy.atlas dcopy.atlas ccopy.atlas zcopy.atlas \
-       sswap.atlas dswap.atlas cswap.atlas zswap.atlas \
-       sscal.atlas dscal.atlas cscal.atlas zscal.atlas \
-       sasum.atlas dasum.atlas casum.atlas zasum.atlas \
-       ssymv.atlas dsymv.atlas csymv.atlas zsymv.atlas \
-       chemv.atlas zhemv.atlas \
-       chemm.acml zhemm.acml \
-       chemm.atlas zhemm.atlas \
-       cherk.atlas zherk.atlas \
-       cher2k.atlas zher2k.atlas \
-       sgemv.atlas dgemv.atlas cgemv.atlas zgemv.atlas \
-       sgeev.atlas dgeev.atlas cgeev.atlas zgeev.atlas \
-       sgesv.atlas dgesv.atlas cgesv.atlas zgesv.atlas \
-       sgetri.atlas dgetri.atlas cgetri.atlas zgetri.atlas \
-       spotrf.atlas dpotrf.atlas cpotrf.atlas zpotrf.atlas \
-       ssymm.atlas dsymm.atlas csymm.atlas zsymm.atlas \
-       isamax.atlas idamax.atlas icamax.atlas izamax.atlas \
-       snrm2.goto dnrm2.goto scnrm2.goto dznrm2.goto
-
-mkl :: slinpack.mkl dlinpack.mkl clinpack.mkl zlinpack.mkl \
-       scholesky.mkl dcholesky.mkl ccholesky.mkl zcholesky.mkl \
-       sgemm.mkl dgemm.mkl cgemm.mkl zgemm.mkl \
-       strmm.mkl dtrmm.mkl ctrmm.mkl ztrmm.mkl \
-       strsm.mkl dtrsm.mkl ctrsm.mkl ztrsm.mkl \
-       ssyr.mkl dsyr.mkl \
-       ssyrk.mkl dsyrk.mkl csyrk.mkl zsyrk.mkl \
-       ssyr2k.mkl dsyr2k.mkl csyr2k.mkl zsyr2k.mkl \
-       sger.mkl dger.mkl cger.mkl zger.mkl \
-       sdot.mkl ddot.mkl cdot.mkl zdot.mkl \
-       saxpy.mkl daxpy.mkl caxpy.mkl zaxpy.mkl \
-       scopy.mkl dcopy.mkl ccopy.mkl zcopy.mkl \
-       sswap.mkl dswap.mkl cswap.mkl zswap.mkl \
-       sscal.mkl dscal.mkl cscal.mkl zscal.mkl \
-       sasum.mkl dasum.mkl casum.mkl zasum.mkl \
-       ssymv.mkl dsymv.mkl csymv.mkl zsymv.mkl \
-       chemv.mkl zhemv.mkl \
-       chemm.mkl zhemm.mkl \
-       cherk.mkl zherk.mkl \
-       cher2k.mkl zher2k.mkl \
-       sgemv.mkl dgemv.mkl cgemv.mkl zgemv.mkl \
-       sgeev.mkl dgeev.mkl cgeev.mkl zgeev.mkl \
-       sgesv.mkl dgesv.mkl cgesv.mkl zgesv.mkl \
-       sgetri.mkl dgetri.mkl cgetri.mkl zgetri.mkl \
-       spotrf.mkl dpotrf.mkl cpotrf.mkl zpotrf.mkl \
-       ssymm.mkl dsymm.mkl csymm.mkl zsymm.mkl
-
-
-
-
-endif
-
-essl :: sgemm.essl strmm.essl dgemm.essl dtrmm.essl  \
-	cgemm.essl ctrmm.essl zgemm.essl ztrmm.essl  \
-	slinpack.essl clinpack.essl dlinpack.essl zlinpack.essl \
-	scholesky.essl ccholesky.essl dcholesky.essl zcholesky.essl \
-	strsm.essl dtrsm.essl ctrsm.essl ztrsm.essl
-
-veclib :: slinpack.veclib dlinpack.veclib clinpack.veclib zlinpack.veclib \
-       scholesky.veclib dcholesky.veclib ccholesky.veclib zcholesky.veclib \
-       sgemm.veclib dgemm.veclib cgemm.veclib zgemm.veclib \
-       strmm.veclib dtrmm.veclib ctrmm.veclib ztrmm.veclib \
-       strsm.veclib dtrsm.veclib ctrsm.veclib ztrsm.veclib \
-       ssyr.veclib dsyr.veclib \
-       ssyrk.veclib dsyrk.veclib csyrk.veclib zsyrk.veclib \
-       ssyr2k.veclib dsyr2k.veclib csyr2k.veclib zsyr2k.veclib \
-       sger.veclib dger.veclib cger.veclib zger.veclib \
-       sdot.veclib ddot.veclib cdot.veclib zdot.veclib \
-       saxpy.veclib daxpy.veclib caxpy.veclib zaxpy.veclib \
-       scopy.veclib dcopy.veclib ccopy.veclib zcopy.veclib \
-       sswap.veclib dswap.veclib cswap.veclib zswap.veclib \
-       sscal.veclib dscal.veclib cscal.veclib zscal.veclib \
-       sasum.veclib dasum.veclib casum.veclib zasum.veclib \
-       ssymv.veclib dsymv.veclib csymv.veclib zsymv.veclib \
-       chemv.veclib zhemv.veclib \
-       chemm.veclib zhemm.veclib \
-       cherk.veclib zherk.veclib \
-       cher2k.veclib zher2k.veclib \
-       sgemv.veclib dgemv.veclib cgemv.veclib zgemv.veclib \
-       sgeev.veclib dgeev.veclib cgeev.veclib zgeev.veclib \
-       sgesv.veclib dgesv.veclib cgesv.veclib zgesv.veclib \
-       sgetri.veclib dgetri.veclib cgetri.veclib zgetri.veclib \
-       spotrf.veclib dpotrf.veclib cpotrf.veclib zpotrf.veclib \
-       ssymm.veclib dsymm.veclib csymm.veclib zsymm.veclib
-
-goto_3m :: cgemm3m.goto zgemm3m.goto
-
-mkl_3m :: cgemm3m.mkl zgemm3m.mkl
-
-all :: goto mkl atlas acml veclib
-
-exe :
-	@./Make_exe.sh
-
-##################################### Slinpack ####################################################
-slinpack.goto : slinpack.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-slinpack.acml : slinpack.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-slinpack.atlas : slinpack.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-slinpack.mkl : slinpack.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-slinpack.veclib : slinpack.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-slinpack.essl : slinpack.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Dlinpack ####################################################
-dlinpack.goto : dlinpack.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-dlinpack.acml : dlinpack.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dlinpack.atlas : dlinpack.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dlinpack.mkl : dlinpack.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dlinpack.veclib : dlinpack.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dlinpack.essl : dlinpack.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Clinpack ####################################################
-
-clinpack.goto : clinpack.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-clinpack.acml : clinpack.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-clinpack.atlas : clinpack.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-clinpack.mkl : clinpack.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-clinpack.veclib : clinpack.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-clinpack.essl : clinpack.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Zlinpack ####################################################
-
-zlinpack.goto : zlinpack.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-zlinpack.acml : zlinpack.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zlinpack.atlas : zlinpack.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zlinpack.mkl : zlinpack.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zlinpack.veclib : zlinpack.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zlinpack.essl : zlinpack.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Scholesky ###################################################
-
-scholesky.goto : scholesky.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-scholesky.acml : scholesky.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-scholesky.atlas : scholesky.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-scholesky.mkl : scholesky.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-scholesky.veclib : scholesky.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-scholesky.essl : scholesky.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Dcholesky ###################################################
-
-dcholesky.goto : dcholesky.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-dcholesky.acml : dcholesky.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dcholesky.atlas : dcholesky.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dcholesky.mkl : dcholesky.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dcholesky.veclib : dcholesky.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dcholesky.essl : dcholesky.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Ccholesky ###################################################
-
-ccholesky.goto : ccholesky.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-ccholesky.acml : ccholesky.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ccholesky.atlas : ccholesky.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ccholesky.mkl : ccholesky.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ccholesky.veclib : ccholesky.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ccholesky.essl : ccholesky.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-
-##################################### Zcholesky ###################################################
-
-zcholesky.goto : zcholesky.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-zcholesky.acml : zcholesky.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zcholesky.atlas : zcholesky.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zcholesky.mkl : zcholesky.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zcholesky.veclib : zcholesky.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zcholesky.essl : zcholesky.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Sgemm ####################################################
-sgemm.goto : sgemm.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-sgemm.acml : sgemm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-sgemm.atlas : sgemm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-sgemm.mkl : sgemm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-sgemm.veclib : sgemm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-sgemm.essl : sgemm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Dgemm ####################################################
-dgemm.goto : dgemm.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-dgemm.acml : dgemm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dgemm.atlas : dgemm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dgemm.mkl : dgemm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dgemm.veclib : dgemm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dgemm.essl : dgemm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Cgemm ####################################################
-
-cgemm.goto : cgemm.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-cgemm.acml : cgemm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cgemm.atlas : cgemm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cgemm.mkl : cgemm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cgemm.veclib : cgemm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cgemm.essl : cgemm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Zgemm ####################################################
-
-zgemm.goto : zgemm.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-zgemm.acml : zgemm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zgemm.atlas : zgemm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zgemm.mkl : zgemm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zgemm.veclib : zgemm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zgemm.essl : zgemm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Ssymm ####################################################
-ssymm.goto : ssymm.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-ssymm.acml : ssymm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ssymm.atlas : ssymm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ssymm.mkl : ssymm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ssymm.veclib : ssymm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Dsymm ####################################################
-dsymm.goto : dsymm.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-dsymm.acml : dsymm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dsymm.atlas : dsymm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dsymm.mkl : dsymm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dsymm.veclib : dsymm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Csymm ####################################################
-
-csymm.goto : csymm.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-csymm.acml : csymm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-csymm.atlas : csymm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-csymm.mkl : csymm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-csymm.veclib : csymm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Zsymm ####################################################
-
-zsymm.goto : zsymm.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-zsymm.acml : zsymm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zsymm.atlas : zsymm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zsymm.mkl : zsymm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zsymm.veclib : zsymm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Strmm ####################################################
-strmm.goto : strmm.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-strmm.acml : strmm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-strmm.atlas : strmm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-strmm.mkl : strmm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-strmm.veclib : strmm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-strmm.essl : strmm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Dtrmm ####################################################
-dtrmm.goto : dtrmm.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-dtrmm.acml : dtrmm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dtrmm.atlas : dtrmm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dtrmm.mkl : dtrmm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dtrmm.veclib : dtrmm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dtrmm.essl : dtrmm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Ctrmm ####################################################
-
-ctrmm.goto : ctrmm.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-ctrmm.acml : ctrmm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ctrmm.atlas : ctrmm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ctrmm.mkl : ctrmm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ctrmm.veclib : ctrmm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ctrmm.essl : ctrmm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Ztrmm ####################################################
-
-ztrmm.goto : ztrmm.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-ztrmm.acml : ztrmm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ztrmm.atlas : ztrmm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ztrmm.mkl : ztrmm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ztrmm.veclib : ztrmm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ztrmm.essl : ztrmm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Strsm ####################################################
-strsm.goto : strsm.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-strsm.acml : strsm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-strsm.atlas : strsm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-strsm.mkl : strsm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-strsm.veclib : strsm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-strsm.essl : strsm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Dtrsm ####################################################
-dtrsm.goto : dtrsm.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-dtrsm.acml : dtrsm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dtrsm.atlas : dtrsm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dtrsm.mkl : dtrsm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dtrsm.veclib : dtrsm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dtrsm.essl : dtrsm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Ctrsm ####################################################
-
-ctrsm.goto : ctrsm.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-ctrsm.acml : ctrsm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ctrsm.atlas : ctrsm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ctrsm.mkl : ctrsm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ctrsm.veclib : ctrsm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ctrsm.essl : ctrsm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Ztrsm ####################################################
-
-ztrsm.goto : ztrsm.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-ztrsm.acml : ztrsm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ztrsm.atlas : ztrsm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ztrsm.mkl : ztrsm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ztrsm.veclib : ztrsm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ztrsm.essl : ztrsm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-##################################### Ssyr ####################################################
-ssyr.goto : ssyr.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-ssyr.acml : ssyr.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ssyr.atlas : ssyr.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ssyr.mkl : ssyr.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ssyr.veclib : ssyr.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-##################################### Dsyr ####################################################
-dsyr.goto : dsyr.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-dsyr.acml : dsyr.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dsyr.atlas : dsyr.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dsyr.mkl : dsyr.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dsyr.veclib : dsyr.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Ssyrk ####################################################
-ssyrk.goto : ssyrk.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-ssyrk.acml : ssyrk.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ssyrk.atlas : ssyrk.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ssyrk.mkl : ssyrk.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ssyrk.veclib : ssyrk.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Dsyrk ####################################################
-dsyrk.goto : dsyrk.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-dsyrk.acml : dsyrk.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dsyrk.atlas : dsyrk.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dsyrk.mkl : dsyrk.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dsyrk.veclib : dsyrk.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Csyrk ####################################################
-
-csyrk.goto : csyrk.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-csyrk.acml : csyrk.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-csyrk.atlas : csyrk.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-csyrk.mkl : csyrk.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-csyrk.veclib : csyrk.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Zsyrk ####################################################
-
-zsyrk.goto : zsyrk.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-zsyrk.acml : zsyrk.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zsyrk.atlas : zsyrk.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zsyrk.mkl : zsyrk.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zsyrk.veclib : zsyrk.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Ssyr2k ####################################################
-ssyr2k.goto : ssyr2k.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-ssyr2k.acml : ssyr2k.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ssyr2k.atlas : ssyr2k.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ssyr2k.mkl : ssyr2k.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ssyr2k.veclib : ssyr2k.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Dsyr2k ####################################################
-dsyr2k.goto : dsyr2k.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-dsyr2k.acml : dsyr2k.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dsyr2k.atlas : dsyr2k.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dsyr2k.mkl : dsyr2k.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dsyr2k.veclib : dsyr2k.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Csyr2k ####################################################
-
-csyr2k.goto : csyr2k.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-csyr2k.acml : csyr2k.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-csyr2k.atlas : csyr2k.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-csyr2k.mkl : csyr2k.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-csyr2k.veclib : csyr2k.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Zsyr2k ####################################################
-
-zsyr2k.goto : zsyr2k.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-zsyr2k.acml : zsyr2k.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zsyr2k.atlas : zsyr2k.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zsyr2k.mkl : zsyr2k.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zsyr2k.veclib : zsyr2k.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Chemm ####################################################
-
-chemm.goto : chemm.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-chemm.acml : chemm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-chemm.atlas : chemm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-chemm.mkl : chemm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-chemm.veclib : chemm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Zhemm ####################################################
-
-zhemm.goto : zhemm.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-zhemm.acml : zhemm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zhemm.atlas : zhemm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zhemm.mkl : zhemm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zhemm.veclib : zhemm.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Cherk ####################################################
-
-cherk.goto : cherk.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-cherk.acml : cherk.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cherk.atlas : cherk.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cherk.mkl : cherk.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cherk.veclib : cherk.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Zherk ####################################################
-
-zherk.goto : zherk.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-zherk.acml : zherk.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zherk.atlas : zherk.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zherk.mkl : zherk.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zherk.veclib : zherk.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Cher2k ####################################################
-
-cher2k.goto : cher2k.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-cher2k.acml : cher2k.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cher2k.atlas : cher2k.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cher2k.mkl : cher2k.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cher2k.veclib : cher2k.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Zher2k ####################################################
-
-zher2k.goto : zher2k.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-zher2k.acml : zher2k.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zher2k.atlas : zher2k.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zher2k.mkl : zher2k.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zher2k.veclib : zher2k.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Sgemv ####################################################
-sgemv.goto : sgemv.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-sgemv.acml : sgemv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-sgemv.atlas : sgemv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-sgemv.mkl : sgemv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-sgemv.veclib : sgemv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Dgemv ####################################################
-dgemv.goto : dgemv.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-dgemv.acml : dgemv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dgemv.atlas : dgemv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dgemv.mkl : dgemv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dgemv.veclib : dgemv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Cgemv ####################################################
-
-cgemv.goto : cgemv.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-cgemv.acml : cgemv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cgemv.atlas : cgemv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cgemv.mkl : cgemv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cgemv.veclib : cgemv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Zgemv ####################################################
-
-zgemv.goto : zgemv.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-zgemv.acml : zgemv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zgemv.atlas : zgemv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zgemv.mkl : zgemv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zgemv.veclib : zgemv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Sger ####################################################
-sger.goto : sger.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-sger.acml : sger.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-sger.atlas : sger.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-sger.mkl : sger.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-sger.veclib : sger.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Dger ####################################################
-dger.goto : dger.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-dger.acml : dger.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dger.atlas : dger.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dger.mkl : dger.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dger.veclib : dger.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Cger ####################################################
-cger.goto : cger.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-cger.acml : cger.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cger.atlas : cger.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cger.mkl : cger.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cger.veclib : cger.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Zger ####################################################
-zger.goto : zger.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-zger.acml : zger.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zger.atlas : zger.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zger.mkl : zger.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zger.veclib : zger.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Ssymv ####################################################
-ssymv.goto : ssymv.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-ssymv.acml : ssymv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ssymv.atlas : ssymv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ssymv.mkl : ssymv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ssymv.veclib : ssymv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Dsymv ####################################################
-dsymv.goto : dsymv.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-dsymv.acml : dsymv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dsymv.atlas : dsymv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dsymv.mkl : dsymv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dsymv.veclib : dsymv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Csymv ####################################################
-csymv.goto : csymv.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-csymv.acml : csymv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-csymv.atlas : csymv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-csymv.mkl : csymv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-csymv.veclib : csymv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Dsymv ####################################################
-zsymv.goto : zsymv.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-zsymv.acml : zsymv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zsymv.atlas : zsymv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zsymv.mkl : zsymv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zsymv.veclib : zsymv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Sgeev ####################################################
-sgeev.goto : sgeev.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-sgeev.acml : sgeev.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-sgeev.atlas : sgeev.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-sgeev.mkl : sgeev.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-sgeev.veclib : sgeev.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Dgeev ####################################################
-dgeev.goto : dgeev.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-dgeev.acml : dgeev.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dgeev.atlas : dgeev.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dgeev.mkl : dgeev.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dgeev.veclib : dgeev.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Cgeev ####################################################
-
-cgeev.goto : cgeev.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-cgeev.acml : cgeev.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cgeev.atlas : cgeev.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cgeev.mkl : cgeev.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cgeev.veclib : cgeev.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Zgeev ####################################################
-
-zgeev.goto : zgeev.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-zgeev.acml : zgeev.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zgeev.atlas : zgeev.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zgeev.mkl : zgeev.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zgeev.veclib : zgeev.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Sgetri ####################################################
-sgetri.goto : sgetri.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-sgetri.acml : sgetri.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-sgetri.atlas : sgetri.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-sgetri.mkl : sgetri.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-sgetri.veclib : sgetri.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Dgetri ####################################################
-dgetri.goto : dgetri.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-dgetri.acml : dgetri.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dgetri.atlas : dgetri.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dgetri.mkl : dgetri.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dgetri.veclib : dgetri.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Cgetri ####################################################
-
-cgetri.goto : cgetri.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-cgetri.acml : cgetri.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cgetri.atlas : cgetri.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cgetri.mkl : cgetri.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cgetri.veclib : cgetri.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Zgetri ####################################################
-
-zgetri.goto : zgetri.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-zgetri.acml : zgetri.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zgetri.atlas : zgetri.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zgetri.mkl : zgetri.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zgetri.veclib : zgetri.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Spotrf ####################################################
-spotrf.goto : spotrf.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-spotrf.acml : spotrf.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-spotrf.atlas : spotrf.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-spotrf.mkl : spotrf.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-spotrf.veclib : spotrf.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Dpotrf ####################################################
-dpotrf.goto : dpotrf.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-dpotrf.acml : dpotrf.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dpotrf.atlas : dpotrf.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dpotrf.mkl : dpotrf.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dpotrf.veclib : dpotrf.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Cpotrf ####################################################
-
-cpotrf.goto : cpotrf.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-cpotrf.acml : cpotrf.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cpotrf.atlas : cpotrf.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cpotrf.mkl : cpotrf.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cpotrf.veclib : cpotrf.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Zpotrf ####################################################
-
-zpotrf.goto : zpotrf.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-zpotrf.acml : zpotrf.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zpotrf.atlas : zpotrf.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zpotrf.mkl : zpotrf.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zpotrf.veclib : zpotrf.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Chemv ####################################################
-
-chemv.goto : chemv.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-chemv.acml : chemv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-chemv.atlas : chemv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-chemv.mkl : chemv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-chemv.veclib : chemv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Zhemv ####################################################
-
-zhemv.goto : zhemv.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-zhemv.acml : zhemv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zhemv.atlas : zhemv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zhemv.mkl : zhemv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zhemv.veclib : zhemv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Sdot ####################################################
-sdot.goto : sdot.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-sdot.acml : sdot.$(SUFFIX)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-sdot.atlas : sdot.$(SUFFIX)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-sdot.mkl : sdot.$(SUFFIX)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-sdot.veclib : sdot.$(SUFFIX)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Ddot ####################################################
-ddot.goto : ddot.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-ddot.acml : ddot.$(SUFFIX)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ddot.atlas : ddot.$(SUFFIX)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ddot.mkl : ddot.$(SUFFIX)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ddot.veclib : ddot.$(SUFFIX)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Cdot ####################################################
-cdot.goto : cdot.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-cdot.acml : cdot.$(SUFFIX)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cdot.atlas : cdot.$(SUFFIX)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cdot.mkl : cdot-intel.$(SUFFIX)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cdot.veclib : cdot-intel.$(SUFFIX)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Zdot ####################################################
-zdot.goto : zdot.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-zdot.acml : zdot.$(SUFFIX)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zdot.atlas : zdot.$(SUFFIX)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zdot.mkl : zdot-intel.$(SUFFIX)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zdot.veclib : zdot-intel.$(SUFFIX)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Srot ####################################################
-srot.goto : srot.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-srot.acml : srot.$(SUFFIX)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-srot.atlas : srot.$(SUFFIX)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-srot.mkl : srot.$(SUFFIX)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-srot.veclib : srot.$(SUFFIX)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Drot ####################################################
-drot.goto : drot.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-drot.acml : drot.$(SUFFIX)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-drot.atlas : drot.$(SUFFIX)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-drot.mkl : drot.$(SUFFIX)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-drot.veclib : drot.$(SUFFIX)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-
-##################################### Saxpy ####################################################
-saxpy.goto : saxpy.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-saxpy.acml : saxpy.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-saxpy.atlas : saxpy.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-saxpy.mkl : saxpy.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-saxpy.veclib : saxpy.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Daxpy ####################################################
-daxpy.goto : daxpy.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-daxpy.acml : daxpy.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-daxpy.atlas : daxpy.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-daxpy.mkl : daxpy.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-daxpy.veclib : daxpy.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Caxpy ####################################################
-
-caxpy.goto : caxpy.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-caxpy.acml : caxpy.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-caxpy.atlas : caxpy.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-caxpy.mkl : caxpy.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-caxpy.veclib : caxpy.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Zaxpy ####################################################
-
-zaxpy.goto : zaxpy.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-zaxpy.acml : zaxpy.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zaxpy.atlas : zaxpy.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zaxpy.mkl : zaxpy.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zaxpy.veclib : zaxpy.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-
-##################################### Scopy ####################################################
-scopy.goto : scopy.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-scopy.acml : scopy.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-scopy.atlas : scopy.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-scopy.mkl : scopy.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-scopy.veclib : scopy.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Dcopy ####################################################
-dcopy.goto : dcopy.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-dcopy.acml : dcopy.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dcopy.atlas : dcopy.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dcopy.mkl : dcopy.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dcopy.veclib : dcopy.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Ccopy ####################################################
-
-ccopy.goto : ccopy.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-ccopy.acml : ccopy.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ccopy.atlas : ccopy.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ccopy.mkl : ccopy.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-ccopy.veclib : ccopy.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Zcopy ####################################################
-
-zcopy.goto : zcopy.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-zcopy.acml : zcopy.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zcopy.atlas : zcopy.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zcopy.mkl : zcopy.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zcopy.veclib : zcopy.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Sscal ####################################################
-sscal.goto : sscal.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-sscal.acml : sscal.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-sscal.atlas : sscal.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-sscal.mkl : sscal.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-sscal.veclib : sscal.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Dscal ####################################################
-dscal.goto : dscal.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-dscal.acml : dscal.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dscal.atlas : dscal.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dscal.mkl : dscal.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dscal.veclib : dscal.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Cscal ####################################################
-
-cscal.goto : cscal.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-cscal.acml : cscal.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cscal.atlas : cscal.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cscal.mkl : cscal.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cscal.veclib : cscal.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Zscal ####################################################
-
-zscal.goto : zscal.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-zscal.acml : zscal.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zscal.atlas : zscal.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zscal.mkl : zscal.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zscal.veclib : zscal.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Sasum ####################################################
-sasum.goto : sasum.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-sasum.acml : sasum.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-sasum.atlas : sasum.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-sasum.mkl : sasum.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-sasum.veclib : sasum.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Dasum ####################################################
-dasum.goto : dasum.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-dasum.acml : dasum.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dasum.atlas : dasum.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dasum.mkl : dasum.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dasum.veclib : dasum.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Casum ####################################################
-
-casum.goto : casum.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-casum.acml : casum.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-casum.atlas : casum.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-casum.mkl : casum.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-casum.veclib : casum.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Zasum ####################################################
-
-zasum.goto : zasum.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-zasum.acml : zasum.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zasum.atlas : zasum.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zasum.mkl : zasum.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zasum.veclib : zasum.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Sswap ####################################################
-sswap.goto : sswap.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-sswap.acml : sswap.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-sswap.atlas : sswap.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-sswap.mkl : sswap.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-sswap.veclib : sswap.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Dswap ####################################################
-dswap.goto : dswap.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-dswap.acml : dswap.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dswap.atlas : dswap.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dswap.mkl : dswap.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dswap.veclib : dswap.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Cswap ####################################################
-
-cswap.goto : cswap.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-cswap.acml : cswap.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cswap.atlas : cswap.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cswap.mkl : cswap.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cswap.veclib : cswap.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Zswap ####################################################
-
-zswap.goto : zswap.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-zswap.acml : zswap.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zswap.atlas : zswap.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zswap.mkl : zswap.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zswap.veclib : zswap.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-
-##################################### Sgesv ####################################################
-sgesv.goto : sgesv.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-sgesv.acml : sgesv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-sgesv.atlas : sgesv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-sgesv.mkl : sgesv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-sgesv.veclib : sgesv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Dgesv ####################################################
-dgesv.goto : dgesv.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-dgesv.acml : dgesv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dgesv.atlas : dgesv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dgesv.mkl : dgesv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-dgesv.veclib : dgesv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Cgesv ####################################################
-
-cgesv.goto : cgesv.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-cgesv.acml : cgesv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cgesv.atlas : cgesv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cgesv.mkl : cgesv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cgesv.veclib : cgesv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Zgesv ####################################################
-
-zgesv.goto : zgesv.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-zgesv.acml : zgesv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zgesv.atlas : zgesv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zgesv.mkl : zgesv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zgesv.veclib : zgesv.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-
-##################################### Cgemm3m ####################################################
-
-cgemm3m.goto : cgemm3m.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-cgemm3m.mkl : cgemm3m.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-cgemm3m.veclib : cgemm3m.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-##################################### Zgemm3m ####################################################
-
-zgemm3m.goto : zgemm3m.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-zgemm3m.mkl : zgemm3m.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-zgemm3m.veclib : zgemm3m.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-############################################## ISAMAX ##############################################
-isamax.goto : isamax.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-isamax.atlas : isamax.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-############################################## IDAMAX ##############################################
-idamax.goto : idamax.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-idamax.atlas : idamax.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-############################################## ICAMAX ##############################################
-icamax.goto : icamax.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-icamax.atlas : icamax.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-############################################## IZAMAX ##############################################
-izamax.goto : izamax.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-izamax.atlas : izamax.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-############################################## SNRM2 ##############################################
-snrm2.goto : snrm2.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-snrm2.atlas : snrm2.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-############################################## DNRM2 ##############################################
-dnrm2.goto : dnrm2.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-dnrm2.atlas : dnrm2.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-############################################## Sscnrm2 ##############################################
-scnrm2.goto : scnrm2.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-scnrm2.atlas : scnrm2.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-############################################## Ddznrm2 ##############################################
-dznrm2.goto : dznrm2.$(SUFFIX) ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
-
-dznrm2.atlas : dznrm2.$(SUFFIX)
-	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
-
-
-###################################################################################################
-
-slinpack.$(SUFFIX) : linpack.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
-
-dlinpack.$(SUFFIX) : linpack.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
-
-clinpack.$(SUFFIX) : linpack.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
-
-zlinpack.$(SUFFIX) : linpack.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
-
-scholesky.$(SUFFIX) : cholesky.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
-
-dcholesky.$(SUFFIX) : cholesky.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
-
-ccholesky.$(SUFFIX) : cholesky.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
-
-zcholesky.$(SUFFIX) : cholesky.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
-
-sgemm.$(SUFFIX) : gemm.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
-
-dgemm.$(SUFFIX) : gemm.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
-
-cgemm.$(SUFFIX) : gemm.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
-
-zgemm.$(SUFFIX) : gemm.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
-
-ssymm.$(SUFFIX) : symm.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
-
-dsymm.$(SUFFIX) : symm.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
-
-csymm.$(SUFFIX) : symm.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
-
-zsymm.$(SUFFIX) : symm.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
-
-strmm.$(SUFFIX) : trmm.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
-
-dtrmm.$(SUFFIX) : trmm.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
-
-ctrmm.$(SUFFIX) : trmm.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
-
-ztrmm.$(SUFFIX) : trmm.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
-
-strsm.$(SUFFIX) : trsm.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
-
-dtrsm.$(SUFFIX) : trsm.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
-
-ctrsm.$(SUFFIX) : trsm.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
-
-ztrsm.$(SUFFIX) : trsm.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
-
-ssyr.$(SUFFIX) : syr.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
-
-dsyr.$(SUFFIX) : syr.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
-
-ssyrk.$(SUFFIX) : syrk.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
-
-dsyrk.$(SUFFIX) : syrk.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
-
-csyrk.$(SUFFIX) : syrk.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
-
-zsyrk.$(SUFFIX) : syrk.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
-
-ssyr2k.$(SUFFIX) : syr2k.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
-
-dsyr2k.$(SUFFIX) : syr2k.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
-
-csyr2k.$(SUFFIX) : syr2k.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
-
-zsyr2k.$(SUFFIX) : syr2k.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
-
-chemm.$(SUFFIX) : hemm.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
-
-zhemm.$(SUFFIX) : hemm.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
-
-cherk.$(SUFFIX) : herk.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
-
-zherk.$(SUFFIX) : herk.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
-
-cher2k.$(SUFFIX) : her2k.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
-
-zher2k.$(SUFFIX) : her2k.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
-
-sgemv.$(SUFFIX) : gemv.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
-
-dgemv.$(SUFFIX) : gemv.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
-
-cgemv.$(SUFFIX) : gemv.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
-
-zgemv.$(SUFFIX) : gemv.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
-
-sger.$(SUFFIX) : ger.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
-
-dger.$(SUFFIX) : ger.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
-
-cger.$(SUFFIX) : ger.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
-
-zger.$(SUFFIX) : ger.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
-
-
-ssymv.$(SUFFIX) : symv.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
-
-dsymv.$(SUFFIX) : symv.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
-
-csymv.$(SUFFIX) : symv.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
-
-zsymv.$(SUFFIX) : symv.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
-
-sgeev.$(SUFFIX) : geev.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
-
-dgeev.$(SUFFIX) : geev.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
-
-cgeev.$(SUFFIX) : geev.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
-
-zgeev.$(SUFFIX) : geev.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
-
-sgetri.$(SUFFIX) : getri.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
-
-dgetri.$(SUFFIX) : getri.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
-
-cgetri.$(SUFFIX) : getri.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
-
-zgetri.$(SUFFIX) : getri.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
-
-spotrf.$(SUFFIX) : potrf.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
-
-dpotrf.$(SUFFIX) : potrf.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
-
-cpotrf.$(SUFFIX) : potrf.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
-
-zpotrf.$(SUFFIX) : potrf.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
-
-chemv.$(SUFFIX) : hemv.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
-
-zhemv.$(SUFFIX) : hemv.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
-
-sdot.$(SUFFIX) : dot.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
-
-ddot.$(SUFFIX) : dot.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
-
-cdot.$(SUFFIX) : zdot.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
-
-zdot.$(SUFFIX) : zdot.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
-
-cdot-intel.$(SUFFIX) : zdot-intel.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
-
-zdot-intel.$(SUFFIX) : zdot-intel.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
-
-
-
-saxpy.$(SUFFIX) : axpy.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
-
-daxpy.$(SUFFIX) : axpy.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
-
-caxpy.$(SUFFIX) : axpy.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
-
-zaxpy.$(SUFFIX) : axpy.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
-
-scopy.$(SUFFIX) : copy.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
-
-dcopy.$(SUFFIX) : copy.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
-
-ccopy.$(SUFFIX) : copy.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
-
-zcopy.$(SUFFIX) : copy.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
-
-sswap.$(SUFFIX) : swap.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
-
-dswap.$(SUFFIX) : swap.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
-
-cswap.$(SUFFIX) : swap.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
-
-zswap.$(SUFFIX) : swap.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
-
-
-
-sscal.$(SUFFIX) : scal.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
-
-dscal.$(SUFFIX) : scal.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
-
-cscal.$(SUFFIX) : scal.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
-
-zscal.$(SUFFIX) : scal.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
-
-sasum.$(SUFFIX) : asum.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
-
-dasum.$(SUFFIX) : asum.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
-
-casum.$(SUFFIX) : asum.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
-
-zasum.$(SUFFIX) : asum.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
-
-
-sgesv.$(SUFFIX) : gesv.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
-
-dgesv.$(SUFFIX) : gesv.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
-
-cgesv.$(SUFFIX) : gesv.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
-
-zgesv.$(SUFFIX) : gesv.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
-
-srot.$(SUFFIX) : rot.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
-
-drot.$(SUFFIX) : rot.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
-
-
-
-
-
-cgemm3m.$(SUFFIX) : gemm3m.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
-
-zgemm3m.$(SUFFIX) : gemm3m.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
-
-
-isamax.$(SUFFIX) : iamax.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
-
-idamax.$(SUFFIX) : iamax.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
-
-icamax.$(SUFFIX) : iamax.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
-
-izamax.$(SUFFIX) : iamax.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
-
-
-snrm2.$(SUFFIX) : nrm2.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
-
-dnrm2.$(SUFFIX) : nrm2.c
-	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
-
-scnrm2.$(SUFFIX) : nrm2.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
-
-dznrm2.$(SUFFIX) : nrm2.c
-	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
-
-
-smallscaling: smallscaling.c ../$(LIBNAME)
-	$(CC) $(CFLAGS) -o $(@F) $^ $(EXTRALIB) -fopenmp -lm -lpthread
-
-clean ::
-	@rm -f *.goto *.mkl *.acml *.atlas *.veclib *.essl smallscaling
-
-include $(TOPDIR)/Makefile.tail
-
+TOPDIR	= ..
+include $(TOPDIR)/Makefile.system
+
+# ACML standard
+#ACML=/opt/acml5.3.1/gfortran64_mp/lib
+#LIBACML = -fopenmp $(ACML)/libacml_mp.a -lgfortran -lm
+
+# ACML custom
+#ACML=/opt/pb/acml-5-3-1-gfortran-64bit/gfortran64_fma4_mp/lib
+#LIBACML = -fopenmp $(ACML)/libacml_mp.a -lgfortran -lm
+
+# ACML 6.1 custom
+ACML=/home/saar/acml6.1/gfortran64_mp/lib
+LIBACML = -fopenmp $(ACML)/libacml_mp.so -lgfortran -lm
+
+
+# Atlas Ubuntu
+#ATLAS=/usr/lib/atlas-base
+#LIBATLAS = -fopenmp $(ATLAS)/liblapack_atlas.a  $(ATLAS)/libptcblas.a  $(ATLAS)/libptf77blas.a  $(ATLAS)/libatlas.a -lgfortran -lm
+
+# Atlas RHEL and Fedora
+ATLAS=/usr/lib64/atlas
+LIBATLAS = -fopenmp $(ATLAS)/liblapack.a  $(ATLAS)/libptcblas.a  $(ATLAS)/libptf77blas.a  $(ATLAS)/libatlas.a -lgfortran -lm
+
+# Intel standard
+# MKL=/opt/intel/mkl/lib/intel64
+# LIBMKL = -L$(MKL) -lmkl_intel_lp64 -lmkl_gnu_thread -lmkl_core -lgomp -lpthread -lm
+
+# Intel custom
+MKL=/home/saar/intel_mkl
+LIBMKL = -L$(MKL) -lmkl_intel_lp64 -lmkl_gnu_thread -lmkl_core -lgomp -lpthread -lm
+
+# Apple vecLib
+LIBVECLIB = -framework Accelerate
+
+ESSL=/opt/ibm/lib
+#LIBESSL = -lesslsmp $(ESSL)/libxlomp_ser.so.1 $(ESSL)/libxlf90_r.so.1 $(ESSL)/libxlfmath.so.1 $(ESSL)/libxlsmp.so.1 /opt/ibm/xlC/13.1.3/lib/libxl.a
+LIBESSL = -lesslsmp  $(ESSL)/libxlf90_r.so.1 $(ESSL)/libxlfmath.so.1 $(ESSL)/libxlsmp.so.1 /opt/ibm/xlC/13.1.3/lib/libxl.a
+
+ifneq ($(NO_LAPACK), 1)
+GOTO_LAPACK_TARGETS=slinpack.goto dlinpack.goto clinpack.goto zlinpack.goto \
+		    scholesky.goto dcholesky.goto ccholesky.goto zcholesky.goto \
+		    sgesv.goto dgesv.goto cgesv.goto zgesv.goto \
+		    sgeev.goto dgeev.goto cgeev.goto zgeev.goto \
+		    csymv.goto zsymv.goto \
+		    sgetri.goto dgetri.goto cgetri.goto zgetri.goto \
+		    spotrf.goto dpotrf.goto cpotrf.goto zpotrf.goto
+else
+GOTO_LAPACK_TARGETS=
+endif
+
+ifeq ($(BUILD_HALF),1)
+GOTO_HALF_TARGETS=shgemm.goto
+else
+GOTO_HALF_TARGETS=
+endif
+
+ifeq ($(OSNAME), WINNT)
+
+goto :: slinpack.goto dlinpack.goto clinpack.goto zlinpack.goto \
+       scholesky.goto dcholesky.goto ccholesky.goto zcholesky.goto \
+       sgemm.goto dgemm.goto cgemm.goto zgemm.goto \
+       strmm.goto dtrmm.goto ctrmm.goto ztrmm.goto \
+       strsm.goto dtrsm.goto ctrsm.goto ztrsm.goto \
+       sspr.goto dspr.goto \
+       sspr2.goto dspr2.goto \
+       ssyr.goto dsyr.goto \
+       ssyr2.goto dsyr2.goto \
+       ssyrk.goto dsyrk.goto csyrk.goto zsyrk.goto \
+       ssyr2k.goto dsyr2k.goto csyr2k.goto zsyr2k.goto \
+       sger.goto dger.goto cger.goto zger.goto \
+       sdot.goto ddot.goto \
+       srot.goto drot.goto csrot.goto zdrot.goto \
+       srotm.goto drotm.goto \
+       saxpy.goto daxpy.goto caxpy.goto zaxpy.goto \
+       scopy.goto dcopy.goto ccopy.goto zcopy.goto \
+       sswap.goto dswap.goto cswap.goto zswap.goto \
+       sscal.goto dscal.goto cscal.goto zscal.goto \
+       sasum.goto dasum.goto casum.goto zasum.goto \
+       ssymv.goto dsymv.goto csymv.goto zsymv.goto \
+       chemv.goto zhemv.goto \
+       chbmv.goto zhbmv.goto \
+       chpmv.goto zhpmv.goto \
+       chemm.goto zhemm.goto \
+       cherk.goto zherk.goto \
+       cher2k.goto zher2k.goto \
+       cher.goto zher.goto \
+       cher2.goto zher2.goto \
+       sgemv.goto dgemv.goto cgemv.goto zgemv.goto \
+	   sspmv.goto dspmv.goto \
+       strmv.goto dtrmv.goto ctrmv.goto ztrmv.goto \
+       stpmv.goto dtpmv.goto ctpmv.goto ztpmv.goto \
+       stpsv.goto dtpsv.goto ctpsv.goto ztpsv.goto \
+       strsv.goto dtrsv.goto ctrsv.goto ztrsv.goto \
+       sgeev.goto dgeev.goto cgeev.goto zgeev.goto \
+       sgesv.goto dgesv.goto cgesv.goto zgesv.goto \
+       sgetri.goto dgetri.goto cgetri.goto zgetri.goto \
+       spotrf.goto dpotrf.goto cpotrf.goto zpotrf.goto \
+       ssymm.goto dsymm.goto csymm.goto zsymm.goto \
+       saxpby.goto daxpby.goto caxpby.goto zaxpby.goto $(GOTO_HALF_TARGETS)
+
+acml :: slinpack.acml dlinpack.acml clinpack.acml zlinpack.acml \
+       scholesky.acml dcholesky.acml ccholesky.acml zcholesky.acml \
+       sgemm.acml dgemm.acml cgemm.acml zgemm.acml \
+       strmm.acml dtrmm.acml ctrmm.acml ztrmm.acml \
+       strsm.acml dtrsm.acml ctrsm.acml ztrsm.acml \
+       sspr.acml dspr.acml \
+       sspr2.acml dspr2.acml \
+       ssyr.acml dsyr.acml \
+       ssyr2.acml dsyr2.acml \
+       ssyrk.acml dsyrk.acml csyrk.acml zsyrk.acml \
+       ssyr2k.acml dsyr2k.acml csyr2k.acml zsyr2k.acml \
+       sger.acml dger.acml cger.acml zger.acml \
+       sdot.acml ddot.acml \
+       srot.acml drot.acml csrot.acml zdrot.acml \
+       srotm.acml drotm.acml \
+       saxpy.acml daxpy.acml caxpy.acml zaxpy.acml \
+       scopy.acml dcopy.acml ccopy.acml zcopy.acml \
+       sswap.acml dswap.acml cswap.acml zswap.acml \
+       sscal.acml dscal.acml cscal.acml zscal.acml \
+       sasum.acml dasum.acml casum.acml zasum.acml \
+       ssymv.acml dsymv.acml csymv.acml zsymv.acml \
+       chemv.acml zhemv.acml \
+       chbmv.acml zhbmv.acml \
+       chpmv.acml zhpmv.acml \
+       chemm.acml zhemm.acml \
+       cherk.acml zherk.acml \
+       cher2k.acml zher2k.acml \
+       cher.acml zher.acml \
+       cher2.acml zher2.acml \
+       sgemv.acml dgemv.acml cgemv.acml zgemv.acml \
+       strmv.acml dtrmv.acml ctrmv.acml ztrmv.acml \
+       stpmv.acml dtpmv.acml ctpmv.acml ztpmv.acml \
+       stpsv.acml dtpsv.acml ctpsv.acml ztpsv.acml \
+       strsv.acml dtrsv.acml ctrsv.acml ztrsv.acml \
+       sgeev.acml dgeev.acml cgeev.acml zgeev.acml \
+       sgesv.acml dgesv.acml cgesv.acml zgesv.acml \
+       sgetri.acml dgetri.acml cgetri.acml zgetri.acml \
+       spotrf.acml dpotrf.acml cpotrf.acml zpotrf.acml \
+       ssymm.acml dsymm.acml csymm.acml zsymm.acml \
+       saxpby.acml daxpby.acml caxpby.acml zaxpby.acml
+
+atlas :: slinpack.atlas dlinpack.atlas clinpack.atlas zlinpack.atlas \
+       scholesky.atlas dcholesky.atlas ccholesky.atlas zcholesky.atlas \
+       sgemm.atlas dgemm.atlas cgemm.atlas zgemm.atlas \
+       strmm.atlas dtrmm.atlas ctrmm.atlas ztrmm.atlas \
+       strsm.atlas dtrsm.atlas ctrsm.atlas ztrsm.atlas \
+       sspr.atlas dspr.atlas \
+       sspr2.atlas dspr2.atlas \
+       ssyr.atlas dsyr.atlas \
+       ssyr2.atlas dsyr2.atlas \
+       ssyrk.atlas dsyrk.atlas csyrk.atlas zsyrk.atlas \
+       ssyr2k.atlas dsyr2k.atlas csyr2k.atlas zsyr2k.atlas \
+       sger.atlas dger.atlas cger.atlas zger.atlas\
+       sdot.atlas ddot.atlas \
+       srot.atlas drot.atlas csrot.atlas zdrot.atlas \
+       srotm.atlas drotm.atlas \
+       saxpy.atlas daxpy.atlas caxpy.atlas zaxpy.atlas \
+       scopy.atlas dcopy.atlas ccopy.atlas zcopy.atlas \
+       sswap.atlas dswap.atlas cswap.atlas zswap.atlas \
+       sscal.atlas dscal.atlas cscal.atlas zscal.atlas \
+       sasum.atlas dasum.atlas casum.atlas zasum.atlas \
+       ssymv.atlas dsymv.atlas csymv.atlas zsymv.atlas \
+       chemv.atlas zhemv.atlas \
+       chbmv.atlas zhbmv.atlas \
+       chpmv.atlas zhpmv.atlas \
+       chemm.acml zhemm.acml \
+       chemm.atlas zhemm.atlas \
+       cherk.atlas zherk.atlas \
+       cher2k.atlas zher2k.atlas \
+       cher.atlas zher.atlas \
+       cher2.atlas zher2.atlas \
+       sgemv.atlas dgemv.atlas cgemv.atlas zgemv.atlas \
+	   sspmv.atlas dspmv.atlas \
+       strmv.atlas dtrmv.atlas ctrmv.atlas ztrmv.atlas \
+       stpmv.atlas dtpmv.atlas ctpmv.atlas ztpmv.atlas \
+       stpsv.atlas dtpsv.atlas ctpsv.atlas ztpsv.atlas \
+       strsv.atlas dtrsv.atlas ctrsv.atlas ztrsv.atlas \
+       sgeev.atlas dgeev.atlas cgeev.atlas zgeev.atlas \
+       sgesv.atlas dgesv.atlas cgesv.atlas zgesv.atlas \
+       sgetri.atlas dgetri.atlas cgetri.atlas zgetri.atlas \
+       spotrf.atlas dpotrf.atlas cpotrf.atlas zpotrf.atlas \
+       ssymm.atlas dsymm.atlas csymm.atlas zsymm.atlas \
+       saxpby.atlas daxpby.atlas caxpby.atlas zaxpby.atlas
+
+mkl :: slinpack.mkl dlinpack.mkl clinpack.mkl zlinpack.mkl \
+       scholesky.mkl dcholesky.mkl ccholesky.mkl zcholesky.mkl \
+       sgemm.mkl dgemm.mkl cgemm.mkl zgemm.mkl \
+       strmm.mkl dtrmm.mkl ctrmm.mkl ztrmm.mkl \
+       strsm.mkl dtrsm.mkl ctrsm.mkl ztrsm.mkl \
+       sspr.mkl dspr.mkl \
+       sspr2.mkl dspr2.mkl \
+       ssyr.mkl dsyr.mkl \
+       ssyr2.mkl dsyr2.mkl \
+       ssyrk.mkl dsyrk.mkl csyrk.mkl zsyrk.mkl \
+       ssyr2k.mkl dsyr2k.mkl csyr2k.mkl zsyr2k.mkl \
+       sger.mkl dger.mkl cger.mkl zger.mkl \
+       sdot.mkl ddot.mkl \
+       srot.mkl drot.mkl csrot.mkl zdrot.mkl \
+       srotm.mkl drotm.mkl \
+       saxpy.mkl daxpy.mkl caxpy.mkl zaxpy.mkl \
+       scopy.mkl dcopy.mkl ccopy.mkl zcopy.mkl \
+       sswap.mkl dswap.mkl cswap.mkl zswap.mkl \
+       sscal.mkl dscal.mkl cscal.mkl zscal.mkl \
+       sasum.mkl dasum.mkl casum.mkl zasum.mkl \
+       ssymv.mkl dsymv.mkl csymv.mkl zsymv.mkl \
+       chemv.mkl zhemv.mkl \
+       chbmv.mkl zhbmv.mkl \
+       chpmv.mkl zhpmv.mkl \
+       chemm.mkl zhemm.mkl \
+       cherk.mkl zherk.mkl \
+       cher2k.mkl zher2k.mkl \
+       cher.mkl zher.mkl \
+       cher2.mkl zher2.mkl \
+       sgemv.mkl dgemv.mkl cgemv.mkl zgemv.mkl \
+       strmv.mkl dtrmv.mkl ctrmv.mkl ztrmv.mkl \
+       stpmv.mkl dtpmv.mkl ctpmv.mkl ztpmv.mkl \
+       stpsv.mkl dtpsv.mkl ctpsv.mkl ztpsv.mkl \
+       strsv.mkl dtrsv.mkl ctrsv.mkl ztrsv.mkl \
+       sgeev.mkl dgeev.mkl cgeev.mkl zgeev.mkl \
+       sgesv.mkl dgesv.mkl cgesv.mkl zgesv.mkl \
+       sgetri.mkl dgetri.mkl cgetri.mkl zgetri.mkl \
+       spotrf.mkl dpotrf.mkl cpotrf.mkl zpotrf.mkl \
+       ssymm.mkl dsymm.mkl csymm.mkl zsymm.mkl \
+       saxpby.mkl daxpby.mkl caxpby.mkl zaxpby.mkl
+
+else
+
+goto :: sgemm.goto dgemm.goto cgemm.goto zgemm.goto \
+       strmm.goto dtrmm.goto ctrmm.goto ztrmm.goto \
+       strsm.goto dtrsm.goto ctrsm.goto ztrsm.goto \
+       sspr.goto dspr.goto \
+       sspr2.goto dspr2.goto \
+       ssyr.goto dsyr.goto \
+       ssyr2.goto dsyr2.goto \
+       ssyrk.goto dsyrk.goto csyrk.goto zsyrk.goto \
+       ssyr2k.goto dsyr2k.goto csyr2k.goto zsyr2k.goto \
+       sger.goto dger.goto cger.goto zger.goto \
+       sdot.goto ddot.goto cdot.goto zdot.goto \
+       srot.goto drot.goto csrot.goto zdrot.goto \
+       srotm.goto drotm.goto \
+       saxpy.goto daxpy.goto caxpy.goto zaxpy.goto \
+       scopy.goto dcopy.goto ccopy.goto zcopy.goto \
+       sswap.goto dswap.goto cswap.goto zswap.goto \
+       sscal.goto dscal.goto cscal.goto zscal.goto \
+       sasum.goto dasum.goto casum.goto zasum.goto \
+       ssymv.goto dsymv.goto \
+       chemv.goto zhemv.goto \
+       chbmv.goto zhbmv.goto \
+       chpmv.goto zhpmv.goto \
+       chemm.goto zhemm.goto \
+       cherk.goto zherk.goto \
+       cher2k.goto zher2k.goto \
+       cher.goto zher.goto \
+       cher2.goto zher2.goto \
+       sgemv.goto dgemv.goto cgemv.goto zgemv.goto \
+	   sspmv.goto dspmv.goto \
+       strmv.goto dtrmv.goto ctrmv.goto ztrmv.goto \
+       stpmv.goto dtpmv.goto ctpmv.goto ztpmv.goto \
+       stpsv.goto dtpsv.goto ctpsv.goto ztpsv.goto \
+       strsv.goto dtrsv.goto ctrsv.goto ztrsv.goto \
+       ssymm.goto dsymm.goto csymm.goto zsymm.goto \
+       smallscaling \
+       isamax.goto idamax.goto icamax.goto izamax.goto \
+       ismax.goto idmax.goto \
+       isamin.goto idamin.goto icamin.goto izamin.goto \
+       ismin.goto idmin.goto \
+       samax.goto damax.goto camax.goto zamax.goto \
+       smax.goto dmax.goto \
+       samin.goto damin.goto camin.goto zamin.goto \
+       smin.goto dmin.goto \
+       saxpby.goto daxpby.goto caxpby.goto zaxpby.goto \
+       snrm2.goto dnrm2.goto scnrm2.goto dznrm2.goto $(GOTO_LAPACK_TARGETS) $(GOTO_HALF_TARGETS)
+
+acml :: slinpack.acml dlinpack.acml clinpack.acml zlinpack.acml \
+       scholesky.acml dcholesky.acml ccholesky.acml zcholesky.acml \
+       sgemm.acml dgemm.acml cgemm.acml zgemm.acml \
+       strmm.acml dtrmm.acml ctrmm.acml ztrmm.acml \
+       strsm.acml dtrsm.acml ctrsm.acml ztrsm.acml \
+       sspr.acml dspr.acml \
+       sspr2.acml dspr2.acml \
+       ssyr.acml dsyr.acml \
+       ssyr2.acml dsyr2.acml \
+       ssyrk.acml dsyrk.acml csyrk.acml zsyrk.acml \
+       ssyr2k.acml dsyr2k.acml csyr2k.acml zsyr2k.acml \
+       sger.acml dger.acml cger.acml zger.acml \
+       sdot.acml ddot.acml \
+       srot.acml drot.acml csrot.acml zdrot.acml \
+       srotm.acml drotm.acml \
+       saxpy.acml daxpy.acml caxpy.acml zaxpy.acml \
+       scopy.acml dcopy.acml ccopy.acml zcopy.acml \
+       sswap.acml dswap.acml cswap.acml zswap.acml \
+       sscal.acml dscal.acml cscal.acml zscal.acml \
+       sasum.acml dasum.acml casum.acml zasum.acml \
+       ssymv.acml dsymv.acml csymv.acml zsymv.acml \
+       chemv.acml zhemv.acml \
+       chbmv.acml zhbmv.acml \
+       chpmv.acml zhpmv.acml \
+       chemm.acml zhemm.acml \
+       cherk.acml zherk.acml \
+       cher2k.acml zher2k.acml \
+       cher.acml zher.acml \
+       cher2.acml zher2.acml \
+       sgemv.acml dgemv.acml cgemv.acml zgemv.acml \
+       strmv.acml dtrmv.acml ctrmv.acml ztrmv.acml \
+       stpmv.acml dtpmv.acml ctpmv.acml ztpmv.acml \
+       stpsv.acml dtpsv.acml ctpsv.acml ztpsv.acml \
+       strsv.acml dtrsv.acml ctrsv.acml ztrsv.acml \
+       sgeev.acml dgeev.acml cgeev.acml zgeev.acml \
+       sgesv.acml dgesv.acml cgesv.acml zgesv.acml \
+       sgetri.acml dgetri.acml cgetri.acml zgetri.acml \
+       spotrf.acml dpotrf.acml cpotrf.acml zpotrf.acml \
+       ssymm.acml dsymm.acml csymm.acml zsymm.acml \
+       saxpby.acml daxpby.acml caxpby.acml zaxpby.acml
+
+atlas :: slinpack.atlas dlinpack.atlas clinpack.atlas zlinpack.atlas \
+       scholesky.atlas dcholesky.atlas ccholesky.atlas zcholesky.atlas \
+       sgemm.atlas dgemm.atlas cgemm.atlas zgemm.atlas \
+       strmm.atlas dtrmm.atlas ctrmm.atlas ztrmm.atlas \
+       strsm.atlas dtrsm.atlas ctrsm.atlas ztrsm.atlas \
+       sspr.atlas dspr.atlas \
+       sspr2.atlas dspr2.atlas \
+       ssyr.atlas dsyr.atlas \
+       ssyr2.atlas dsyr2.atlas \
+       ssyrk.atlas dsyrk.atlas csyrk.atlas zsyrk.atlas \
+       ssyr2k.atlas dsyr2k.atlas csyr2k.atlas zsyr2k.atlas \
+       sger.atlas dger.atlas cger.atlas zger.atlas\
+       sdot.atlas ddot.atlas \
+       srot.atlas drot.atlas csrot.atlas zdrot.atlas \
+       srotm.atlas drotm.atlas \
+       saxpy.atlas daxpy.atlas caxpy.atlas zaxpy.atlas \
+       scopy.atlas dcopy.atlas ccopy.atlas zcopy.atlas \
+       sswap.atlas dswap.atlas cswap.atlas zswap.atlas \
+       sscal.atlas dscal.atlas cscal.atlas zscal.atlas \
+       sasum.atlas dasum.atlas casum.atlas zasum.atlas \
+       ssymv.atlas dsymv.atlas csymv.atlas zsymv.atlas \
+       chemv.atlas zhemv.atlas \
+       chbmv.atlas zhbmv.atlas \
+       chpmv.atlas zhpmv.atlas \
+       chemm.acml zhemm.acml \
+       chemm.atlas zhemm.atlas \
+       cherk.atlas zherk.atlas \
+       cher2k.atlas zher2k.atlas \
+       cher.atlas zher.atlas \
+       cher2.atlas zher2.atlas \
+       sgemv.atlas dgemv.atlas cgemv.atlas zgemv.atlas \
+	   sspmv.atlas dspmv.atlas \
+       strmv.atlas dtrmv.atlas ctrmv.atlas ztrmv.atlas \
+       stpmv.atlas dtpmv.atlas ctpmv.atlas ztpmv.atlas \
+       stpsv.atlas dtpsv.atlas ctpsv.atlas ztpsv.atlas \
+       strsv.atlas dtrsv.atlas ctrsv.atlas ztrsv.atlas \
+       sgeev.atlas dgeev.atlas cgeev.atlas zgeev.atlas \
+       sgesv.atlas dgesv.atlas cgesv.atlas zgesv.atlas \
+       sgetri.atlas dgetri.atlas cgetri.atlas zgetri.atlas \
+       spotrf.atlas dpotrf.atlas cpotrf.atlas zpotrf.atlas \
+       ssymm.atlas dsymm.atlas csymm.atlas zsymm.atlas \
+       isamax.atlas idamax.atlas icamax.atlas izamax.atlas \
+       snrm2.atlas dnrm2.atlas scnrm2.atlas dznrm2.atlas \
+       saxpby.atlas daxpby.atlas caxpby.atlas zaxpby.atlas
+
+mkl :: slinpack.mkl dlinpack.mkl clinpack.mkl zlinpack.mkl \
+       scholesky.mkl dcholesky.mkl ccholesky.mkl zcholesky.mkl \
+       sgemm.mkl dgemm.mkl cgemm.mkl zgemm.mkl \
+       strmm.mkl dtrmm.mkl ctrmm.mkl ztrmm.mkl \
+       strsm.mkl dtrsm.mkl ctrsm.mkl ztrsm.mkl \
+       sspr.mkl dspr.mkl \
+       sspr2.mkl dspr2.mkl \
+       ssyr.mkl dsyr.mkl \
+       ssyr2.mkl dsyr2.mkl \
+       ssyrk.mkl dsyrk.mkl csyrk.mkl zsyrk.mkl \
+       ssyr2k.mkl dsyr2k.mkl csyr2k.mkl zsyr2k.mkl \
+       sger.mkl dger.mkl cger.mkl zger.mkl \
+       sdot.mkl ddot.mkl cdot.mkl zdot.mkl \
+       srot.atlas drot.atlas csrot.atlas zdrot.atlas \
+       srotm.atlas drotm.atlas \
+       saxpy.mkl daxpy.mkl caxpy.mkl zaxpy.mkl \
+       scopy.mkl dcopy.mkl ccopy.mkl zcopy.mkl \
+       sswap.mkl dswap.mkl cswap.mkl zswap.mkl \
+       sscal.mkl dscal.mkl cscal.mkl zscal.mkl \
+       sasum.mkl dasum.mkl casum.mkl zasum.mkl \
+       ssymv.mkl dsymv.mkl csymv.mkl zsymv.mkl \
+       chemv.mkl zhemv.mkl \
+       chbmv.mkl zhbmv.mkl \
+       chpmv.mkl zhpmv.mkl \
+       chemm.mkl zhemm.mkl \
+       cherk.mkl zherk.mkl \
+       cher2k.mkl zher2k.mkl \
+       cher.mkl zher.mkl \
+       cher2.mkl zher2.mkl \
+       sgemv.mkl dgemv.mkl cgemv.mkl zgemv.mkl \
+       strmv.mkl dtrmv.mkl ctrmv.mkl ztrmv.mkl \
+       stpmv.mkl dtpmv.mkl ctpmv.mkl ztpmv.mkl \
+       stpsv.mkl dtpsv.mkl ctpsv.mkl ztpsv.mkl \
+       strsv.mkl dtrsv.mkl ctrsv.mkl ztrsv.mkl \
+       sgeev.mkl dgeev.mkl cgeev.mkl zgeev.mkl \
+       sgesv.mkl dgesv.mkl cgesv.mkl zgesv.mkl \
+       sgetri.mkl dgetri.mkl cgetri.mkl zgetri.mkl \
+       spotrf.mkl dpotrf.mkl cpotrf.mkl zpotrf.mkl \
+       ssymm.mkl dsymm.mkl csymm.mkl zsymm.mkl \
+       saxpby.mkl daxpby.mkl caxpby.mkl zaxpby.mkl
+
+
+
+
+endif
+
+essl :: sgemm.essl strmm.essl dgemm.essl dtrmm.essl  \
+	cgemm.essl ctrmm.essl zgemm.essl ztrmm.essl  \
+	slinpack.essl clinpack.essl dlinpack.essl zlinpack.essl \
+	scholesky.essl ccholesky.essl dcholesky.essl zcholesky.essl \
+	strsm.essl dtrsm.essl ctrsm.essl ztrsm.essl
+
+veclib :: slinpack.veclib dlinpack.veclib clinpack.veclib zlinpack.veclib \
+       scholesky.veclib dcholesky.veclib ccholesky.veclib zcholesky.veclib \
+       sgemm.veclib dgemm.veclib cgemm.veclib zgemm.veclib \
+       strmm.veclib dtrmm.veclib ctrmm.veclib ztrmm.veclib \
+       strsm.veclib dtrsm.veclib ctrsm.veclib ztrsm.veclib \
+       sspr.veclib dspr.veclib \
+       sspr2.veclib dspr2.veclib \
+       ssyr.veclib dsyr.veclib \
+       ssyr2.veclib dsyr2.veclib \
+       ssyrk.veclib dsyrk.veclib csyrk.veclib zsyrk.veclib \
+       ssyr2k.veclib dsyr2k.veclib csyr2k.veclib zsyr2k.veclib \
+       sger.veclib dger.veclib cger.veclib zger.veclib \
+       sdot.veclib ddot.veclib cdot.veclib zdot.veclib \
+       srot.veclib drot.veclib csrot.veclib zdrot.veclib \
+       srotm.veclib drotm.veclib \
+       saxpy.veclib daxpy.veclib caxpy.veclib zaxpy.veclib \
+       scopy.veclib dcopy.veclib ccopy.veclib zcopy.veclib \
+       sswap.veclib dswap.veclib cswap.veclib zswap.veclib \
+       sscal.veclib dscal.veclib cscal.veclib zscal.veclib \
+       sasum.veclib dasum.veclib casum.veclib zasum.veclib \
+       ssymv.veclib dsymv.veclib csymv.veclib zsymv.veclib \
+       chemv.veclib zhemv.veclib \
+       chbmv.veclib zhbmv.veclib \
+       chpmv.veclib zhpmv.veclib \
+       chemm.veclib zhemm.veclib \
+       cherk.veclib zherk.veclib \
+       cher2k.veclib zher2k.veclib \
+       cher.veclib zher.veclib \
+       cher2.veclib zher2.veclib \
+       sgemv.veclib dgemv.veclib cgemv.veclib zgemv.veclib \
+       strmv.veclib dtrmv.veclib ctrmv.veclib ztrmv.veclib \
+       stpmv.veclib dtpmv.veclib ctpmv.veclib ztpmv.veclib \
+       stpsv.veclib dtpsv.veclib ctpsv.veclib ztpsv.veclib \
+       strsv.veclib dtrsv.veclib ctrsv.veclib ztrsv.veclib \
+       sgeev.veclib dgeev.veclib cgeev.veclib zgeev.veclib \
+       sgesv.veclib dgesv.veclib cgesv.veclib zgesv.veclib \
+       sgetri.veclib dgetri.veclib cgetri.veclib zgetri.veclib \
+       spotrf.veclib dpotrf.veclib cpotrf.veclib zpotrf.veclib \
+       ssymm.veclib dsymm.veclib csymm.veclib zsymm.veclib \
+       saxpby.veclib daxpby.veclib caxpby.veclib zaxpby.veclib
+
+goto_3m :: cgemm3m.goto zgemm3m.goto
+
+mkl_3m :: cgemm3m.mkl zgemm3m.mkl
+
+all :: goto mkl atlas acml veclib
+
+exe :
+	@./Make_exe.sh
+
+##################################### Slinpack ####################################################
+slinpack.goto : slinpack.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+slinpack.acml : slinpack.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+slinpack.atlas : slinpack.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+slinpack.mkl : slinpack.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+slinpack.veclib : slinpack.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+slinpack.essl : slinpack.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Dlinpack ####################################################
+dlinpack.goto : dlinpack.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+dlinpack.acml : dlinpack.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dlinpack.atlas : dlinpack.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dlinpack.mkl : dlinpack.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dlinpack.veclib : dlinpack.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dlinpack.essl : dlinpack.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Clinpack ####################################################
+
+clinpack.goto : clinpack.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+clinpack.acml : clinpack.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+clinpack.atlas : clinpack.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+clinpack.mkl : clinpack.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+clinpack.veclib : clinpack.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+clinpack.essl : clinpack.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Zlinpack ####################################################
+
+zlinpack.goto : zlinpack.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+zlinpack.acml : zlinpack.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zlinpack.atlas : zlinpack.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zlinpack.mkl : zlinpack.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zlinpack.veclib : zlinpack.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zlinpack.essl : zlinpack.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Scholesky ###################################################
+
+scholesky.goto : scholesky.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+scholesky.acml : scholesky.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+scholesky.atlas : scholesky.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+scholesky.mkl : scholesky.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+scholesky.veclib : scholesky.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+scholesky.essl : scholesky.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Dcholesky ###################################################
+
+dcholesky.goto : dcholesky.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+dcholesky.acml : dcholesky.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dcholesky.atlas : dcholesky.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dcholesky.mkl : dcholesky.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dcholesky.veclib : dcholesky.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dcholesky.essl : dcholesky.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Ccholesky ###################################################
+
+ccholesky.goto : ccholesky.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+ccholesky.acml : ccholesky.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ccholesky.atlas : ccholesky.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ccholesky.mkl : ccholesky.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ccholesky.veclib : ccholesky.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ccholesky.essl : ccholesky.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+
+##################################### Zcholesky ###################################################
+
+zcholesky.goto : zcholesky.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+zcholesky.acml : zcholesky.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zcholesky.atlas : zcholesky.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zcholesky.mkl : zcholesky.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zcholesky.veclib : zcholesky.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zcholesky.essl : zcholesky.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Sgemm ####################################################
+ifeq ($(BUILD_HALF),1)
+shgemm.goto : shgemm.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+endif
+
+sgemm.goto : sgemm.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+sgemm.acml : sgemm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sgemm.atlas : sgemm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sgemm.mkl : sgemm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sgemm.veclib : sgemm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sgemm.essl : sgemm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Dgemm ####################################################
+dgemm.goto : dgemm.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+dgemm.acml : dgemm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dgemm.atlas : dgemm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dgemm.mkl : dgemm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dgemm.veclib : dgemm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dgemm.essl : dgemm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Cgemm ####################################################
+
+cgemm.goto : cgemm.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+cgemm.acml : cgemm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cgemm.atlas : cgemm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cgemm.mkl : cgemm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cgemm.veclib : cgemm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cgemm.essl : cgemm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Zgemm ####################################################
+
+zgemm.goto : zgemm.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+zgemm.acml : zgemm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zgemm.atlas : zgemm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zgemm.mkl : zgemm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zgemm.veclib : zgemm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zgemm.essl : zgemm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Ssymm ####################################################
+ssymm.goto : ssymm.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+ssymm.acml : ssymm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ssymm.atlas : ssymm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ssymm.mkl : ssymm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ssymm.veclib : ssymm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Dsymm ####################################################
+dsymm.goto : dsymm.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+dsymm.acml : dsymm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dsymm.atlas : dsymm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dsymm.mkl : dsymm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dsymm.veclib : dsymm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Csymm ####################################################
+
+csymm.goto : csymm.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+csymm.acml : csymm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+csymm.atlas : csymm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+csymm.mkl : csymm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+csymm.veclib : csymm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Zsymm ####################################################
+
+zsymm.goto : zsymm.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+zsymm.acml : zsymm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zsymm.atlas : zsymm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zsymm.mkl : zsymm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zsymm.veclib : zsymm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Strmm ####################################################
+strmm.goto : strmm.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+strmm.acml : strmm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+strmm.atlas : strmm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+strmm.mkl : strmm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+strmm.veclib : strmm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+strmm.essl : strmm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Dtrmm ####################################################
+dtrmm.goto : dtrmm.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+dtrmm.acml : dtrmm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dtrmm.atlas : dtrmm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dtrmm.mkl : dtrmm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dtrmm.veclib : dtrmm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dtrmm.essl : dtrmm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Ctrmm ####################################################
+
+ctrmm.goto : ctrmm.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+ctrmm.acml : ctrmm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ctrmm.atlas : ctrmm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ctrmm.mkl : ctrmm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ctrmm.veclib : ctrmm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ctrmm.essl : ctrmm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Ztrmm ####################################################
+
+ztrmm.goto : ztrmm.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+ztrmm.acml : ztrmm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ztrmm.atlas : ztrmm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ztrmm.mkl : ztrmm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ztrmm.veclib : ztrmm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ztrmm.essl : ztrmm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Strsm ####################################################
+strsm.goto : strsm.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+strsm.acml : strsm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+strsm.atlas : strsm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+strsm.mkl : strsm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+strsm.veclib : strsm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+strsm.essl : strsm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Dtrsm ####################################################
+dtrsm.goto : dtrsm.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+dtrsm.acml : dtrsm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dtrsm.atlas : dtrsm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dtrsm.mkl : dtrsm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dtrsm.veclib : dtrsm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dtrsm.essl : dtrsm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Ctrsm ####################################################
+
+ctrsm.goto : ctrsm.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+ctrsm.acml : ctrsm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ctrsm.atlas : ctrsm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ctrsm.mkl : ctrsm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ctrsm.veclib : ctrsm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ctrsm.essl : ctrsm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Ztrsm ####################################################
+
+ztrsm.goto : ztrsm.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+ztrsm.acml : ztrsm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ztrsm.atlas : ztrsm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ztrsm.mkl : ztrsm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ztrsm.veclib : ztrsm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ztrsm.essl : ztrsm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+##################################### Ssyr ####################################################
+ssyr.goto : ssyr.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+ssyr.acml : ssyr.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ssyr.atlas : ssyr.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ssyr.mkl : ssyr.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ssyr.veclib : ssyr.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+##################################### Dsyr ####################################################
+dsyr.goto : dsyr.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+dsyr.acml : dsyr.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dsyr.atlas : dsyr.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dsyr.mkl : dsyr.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dsyr.veclib : dsyr.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+	
+##################################### Sspr ####################################################
+sspr.goto : sspr.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+sspr.acml : sspr.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sspr.atlas : sspr.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sspr.mkl : sspr.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sspr.veclib : sspr.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+	
+##################################### Dspr ####################################################
+dspr.goto : dspr.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+dspr.acml : dspr.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dspr.atlas : dspr.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dspr.mkl : dspr.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dspr.veclib : dspr.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+	
+##################################### Sspr2 ####################################################
+sspr2.goto : sspr2.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+sspr2.acml : sspr2.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sspr2.atlas : sspr2.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sspr2.mkl : sspr2.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sspr2.veclib : sspr2.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+	
+##################################### Dspr2 ####################################################
+dspr2.goto : dspr2.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+dspr2.acml : dspr2.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dspr2.atlas : dspr2.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dspr2.mkl : dspr2.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dspr2.veclib : dspr2.$(SUFFIX)
+
+##################################### Ssyr2 ####################################################
+ssyr2.goto : ssyr2.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+ssyr2.acml : ssyr2.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ssyr2.atlas : ssyr2.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ssyr2.mkl : ssyr2.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ssyr2.veclib : ssyr2.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+##################################### Dsyr2 ####################################################
+dsyr2.goto : dsyr2.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+dsyr2.acml : dsyr2.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dsyr2.atlas : dsyr2.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dsyr2.mkl : dsyr2.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dsyr2.veclib : dsyr2.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Ssyrk ####################################################
+ssyrk.goto : ssyrk.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+ssyrk.acml : ssyrk.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ssyrk.atlas : ssyrk.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ssyrk.mkl : ssyrk.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ssyrk.veclib : ssyrk.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Dsyrk ####################################################
+dsyrk.goto : dsyrk.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+dsyrk.acml : dsyrk.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dsyrk.atlas : dsyrk.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dsyrk.mkl : dsyrk.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dsyrk.veclib : dsyrk.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Csyrk ####################################################
+
+csyrk.goto : csyrk.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+csyrk.acml : csyrk.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+csyrk.atlas : csyrk.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+csyrk.mkl : csyrk.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+csyrk.veclib : csyrk.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Zsyrk ####################################################
+
+zsyrk.goto : zsyrk.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+zsyrk.acml : zsyrk.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zsyrk.atlas : zsyrk.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zsyrk.mkl : zsyrk.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zsyrk.veclib : zsyrk.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Ssyr2k ####################################################
+ssyr2k.goto : ssyr2k.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+ssyr2k.acml : ssyr2k.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ssyr2k.atlas : ssyr2k.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ssyr2k.mkl : ssyr2k.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ssyr2k.veclib : ssyr2k.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Dsyr2k ####################################################
+dsyr2k.goto : dsyr2k.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+dsyr2k.acml : dsyr2k.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dsyr2k.atlas : dsyr2k.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dsyr2k.mkl : dsyr2k.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dsyr2k.veclib : dsyr2k.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Csyr2k ####################################################
+
+csyr2k.goto : csyr2k.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+csyr2k.acml : csyr2k.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+csyr2k.atlas : csyr2k.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+csyr2k.mkl : csyr2k.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+csyr2k.veclib : csyr2k.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Zsyr2k ####################################################
+
+zsyr2k.goto : zsyr2k.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+zsyr2k.acml : zsyr2k.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zsyr2k.atlas : zsyr2k.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zsyr2k.mkl : zsyr2k.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zsyr2k.veclib : zsyr2k.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Chemm ####################################################
+
+chemm.goto : chemm.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+chemm.acml : chemm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+chemm.atlas : chemm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+chemm.mkl : chemm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+chemm.veclib : chemm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Zhemm ####################################################
+
+zhemm.goto : zhemm.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+zhemm.acml : zhemm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zhemm.atlas : zhemm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zhemm.mkl : zhemm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zhemm.veclib : zhemm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Cherk ####################################################
+
+cherk.goto : cherk.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+cherk.acml : cherk.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cherk.atlas : cherk.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cherk.mkl : cherk.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cherk.veclib : cherk.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Zherk ####################################################
+
+zherk.goto : zherk.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+zherk.acml : zherk.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zherk.atlas : zherk.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zherk.mkl : zherk.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zherk.veclib : zherk.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Cher2k ####################################################
+
+cher2k.goto : cher2k.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+cher2k.acml : cher2k.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cher2k.atlas : cher2k.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cher2k.mkl : cher2k.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cher2k.veclib : cher2k.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Zher2k ####################################################
+
+zher2k.goto : zher2k.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+zher2k.acml : zher2k.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zher2k.atlas : zher2k.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zher2k.mkl : zher2k.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zher2k.veclib : zher2k.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Cher ####################################################
+
+cher.goto : cher.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+cher.acml : cher.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cher.atlas : cher.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cher.mkl : cher.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cher.veclib : cher.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Zher ####################################################
+
+zher.goto : zher.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+zher.acml : zher.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zher.atlas : zher.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zher.mkl : zher.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zher.veclib : zher.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Cher2 ####################################################
+
+cher2.goto : cher2.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+cher2.acml : cher2.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cher2.atlas : cher2.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cher2.mkl : cher2.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cher2.veclib : cher2.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Zher2 ####################################################
+
+zher2.goto : zher2.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+zher2.acml : zher2.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zher2.atlas : zher2.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zher2.mkl : zher2.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zher2.veclib : zher2.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Sgemv ####################################################
+sgemv.goto : sgemv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+sgemv.acml : sgemv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sgemv.atlas : sgemv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sgemv.mkl : sgemv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sgemv.veclib : sgemv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Dgemv ####################################################
+dgemv.goto : dgemv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+dgemv.acml : dgemv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dgemv.atlas : dgemv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dgemv.mkl : dgemv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dgemv.veclib : dgemv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Cgemv ####################################################
+
+cgemv.goto : cgemv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+cgemv.acml : cgemv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cgemv.atlas : cgemv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cgemv.mkl : cgemv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cgemv.veclib : cgemv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Zgemv ####################################################
+
+zgemv.goto : zgemv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+zgemv.acml : zgemv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zgemv.atlas : zgemv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zgemv.mkl : zgemv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zgemv.veclib : zgemv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Sspmv ####################################################
+sspmv.goto : sspmv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+sspmv.atlas : sspmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Dspmv ####################################################
+dspmv.goto : dspmv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+dspmv.atlas : dspmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Strmv ####################################################
+strmv.goto : strmv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+strmv.acml : strmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+strmv.atlas : strmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+strmv.mkl : strmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+strmv.veclib : strmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Dtrmv ####################################################
+dtrmv.goto : dtrmv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+dtrmv.acml : dtrmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dtrmv.atlas : dtrmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dtrmv.mkl : dtrmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dtrmv.veclib : dtrmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Ctrmv ####################################################
+
+ctrmv.goto : ctrmv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+ctrmv.acml : ctrmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ctrmv.atlas : ctrmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ctrmv.mkl : ctrmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ctrmv.veclib : ctrmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Ztrmv ####################################################
+
+ztrmv.goto : ztrmv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+ztrmv.acml : ztrmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ztrmv.atlas : ztrmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ztrmv.mkl : ztrmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ztrmv.veclib : ztrmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+
+##################################### Stpmv ####################################################
+stpmv.goto : stpmv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+stpmv.acml : stpmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+stpmv.atlas : stpmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+stpmv.mkl : stpmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+stpmv.veclib : stpmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Dtpmv ####################################################
+dtpmv.goto : dtpmv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+dtpmv.acml : dtpmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dtpmv.atlas : dtpmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dtpmv.mkl : dtpmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dtpmv.veclib : dtpmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Ctpmv ####################################################
+
+ctpmv.goto : ctpmv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+ctpmv.acml : ctpmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ctpmv.atlas : ctpmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ctpmv.mkl : ctpmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ctpmv.veclib : ctpmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Ztpmv ####################################################
+
+ztpmv.goto : ztpmv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+ztpmv.acml : ztpmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ztpmv.atlas : ztpmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ztpmv.mkl : ztpmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ztpmv.veclib : ztpmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Stpsv ####################################################
+stpsv.goto : stpsv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+stpsv.acml : stpsv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+stpsv.atlas : stpsv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+stpsv.mkl : stpsv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+stpsv.veclib : stpsv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Dtpsv ####################################################
+dtpsv.goto : dtpsv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+dtpsv.acml : dtpsv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dtpsv.atlas : dtpsv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dtpsv.mkl : dtpsv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dtpsv.veclib : dtpsv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Ctpsv ####################################################
+
+ctpsv.goto : ctpsv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+ctpsv.acml : ctpsv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ctpsv.atlas : ctpsv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ctpsv.mkl : ctpsv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ctpsv.veclib : ctpsv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Ztpsv ####################################################
+
+ztpsv.goto : ztpsv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+ztpsv.acml : ztpsv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ztpsv.atlas : ztpsv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ztpsv.mkl : ztpsv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ztpsv.veclib : ztpsv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Strsv ####################################################
+strsv.goto : strsv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+strsv.acml : strsv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+strsv.atlas : strsv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+strsv.mkl : strsv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+strsv.veclib : strsv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Dtrsv ####################################################
+dtrsv.goto : dtrsv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+dtrsv.acml : dtrsv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dtrsv.atlas : dtrsv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dtrsv.mkl : dtrsv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dtrsv.veclib : dtrsv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Ctrsv ####################################################
+
+ctrsv.goto : ctrsv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+ctrsv.acml : ctrsv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ctrsv.atlas : ctrsv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ctrsv.mkl : ctrsv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ctrsv.veclib : ctrsv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Ztrsv ####################################################
+
+ztrsv.goto : ztrsv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+ztrsv.acml : ztrsv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ztrsv.atlas : ztrsv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ztrsv.mkl : ztrsv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ztrsv.veclib : ztrsv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Sger ####################################################
+sger.goto : sger.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+sger.acml : sger.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sger.atlas : sger.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sger.mkl : sger.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sger.veclib : sger.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Dger ####################################################
+dger.goto : dger.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+dger.acml : dger.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dger.atlas : dger.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dger.mkl : dger.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dger.veclib : dger.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Cger ####################################################
+cger.goto : cger.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+cger.acml : cger.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cger.atlas : cger.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cger.mkl : cger.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cger.veclib : cger.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Zger ####################################################
+zger.goto : zger.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+zger.acml : zger.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zger.atlas : zger.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zger.mkl : zger.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zger.veclib : zger.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Ssymv ####################################################
+ssymv.goto : ssymv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+ssymv.acml : ssymv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ssymv.atlas : ssymv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ssymv.mkl : ssymv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ssymv.veclib : ssymv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Dsymv ####################################################
+dsymv.goto : dsymv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+dsymv.acml : dsymv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dsymv.atlas : dsymv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dsymv.mkl : dsymv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dsymv.veclib : dsymv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Csymv ####################################################
+csymv.goto : csymv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+csymv.acml : csymv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+csymv.atlas : csymv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+csymv.mkl : csymv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+csymv.veclib : csymv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Dsymv ####################################################
+zsymv.goto : zsymv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+zsymv.acml : zsymv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zsymv.atlas : zsymv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zsymv.mkl : zsymv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zsymv.veclib : zsymv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Sgeev ####################################################
+sgeev.goto : sgeev.$(SUFFIX) ../$(LIBNAME)
+	$(FC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+sgeev.acml : sgeev.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sgeev.atlas : sgeev.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sgeev.mkl : sgeev.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sgeev.veclib : sgeev.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Dgeev ####################################################
+dgeev.goto : dgeev.$(SUFFIX) ../$(LIBNAME)
+	$(FC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+dgeev.acml : dgeev.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dgeev.atlas : dgeev.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dgeev.mkl : dgeev.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dgeev.veclib : dgeev.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Cgeev ####################################################
+
+cgeev.goto : cgeev.$(SUFFIX) ../$(LIBNAME)
+	$(FC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+cgeev.acml : cgeev.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cgeev.atlas : cgeev.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cgeev.mkl : cgeev.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cgeev.veclib : cgeev.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Zgeev ####################################################
+
+zgeev.goto : zgeev.$(SUFFIX) ../$(LIBNAME)
+	$(FC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+zgeev.acml : zgeev.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zgeev.atlas : zgeev.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zgeev.mkl : zgeev.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zgeev.veclib : zgeev.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Sgetri ####################################################
+sgetri.goto : sgetri.$(SUFFIX) ../$(LIBNAME)
+	$(FC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+sgetri.acml : sgetri.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sgetri.atlas : sgetri.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sgetri.mkl : sgetri.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sgetri.veclib : sgetri.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Dgetri ####################################################
+dgetri.goto : dgetri.$(SUFFIX) ../$(LIBNAME)
+	$(FC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+dgetri.acml : dgetri.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dgetri.atlas : dgetri.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dgetri.mkl : dgetri.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dgetri.veclib : dgetri.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Cgetri ####################################################
+
+cgetri.goto : cgetri.$(SUFFIX) ../$(LIBNAME)
+	$(FC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+cgetri.acml : cgetri.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cgetri.atlas : cgetri.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cgetri.mkl : cgetri.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cgetri.veclib : cgetri.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Zgetri ####################################################
+
+zgetri.goto : zgetri.$(SUFFIX) ../$(LIBNAME)
+	$(FC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+zgetri.acml : zgetri.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zgetri.atlas : zgetri.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zgetri.mkl : zgetri.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zgetri.veclib : zgetri.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Spotrf ####################################################
+spotrf.goto : spotrf.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+spotrf.acml : spotrf.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+spotrf.atlas : spotrf.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+spotrf.mkl : spotrf.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+spotrf.veclib : spotrf.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Dpotrf ####################################################
+dpotrf.goto : dpotrf.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+dpotrf.acml : dpotrf.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dpotrf.atlas : dpotrf.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dpotrf.mkl : dpotrf.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dpotrf.veclib : dpotrf.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Cpotrf ####################################################
+
+cpotrf.goto : cpotrf.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+cpotrf.acml : cpotrf.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cpotrf.atlas : cpotrf.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cpotrf.mkl : cpotrf.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cpotrf.veclib : cpotrf.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Zpotrf ####################################################
+
+zpotrf.goto : zpotrf.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+zpotrf.acml : zpotrf.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zpotrf.atlas : zpotrf.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zpotrf.mkl : zpotrf.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zpotrf.veclib : zpotrf.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Chemv ####################################################
+
+chemv.goto : chemv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+chemv.acml : chemv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+chemv.atlas : chemv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+chemv.mkl : chemv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+chemv.veclib : chemv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Zhemv ####################################################
+
+zhemv.goto : zhemv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+zhemv.acml : zhemv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zhemv.atlas : zhemv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zhemv.mkl : zhemv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zhemv.veclib : zhemv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+##################################### Chbmv ####################################################
+
+chbmv.goto : chbmv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+chbmv.acml : chbmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+chbmv.atlas : chbmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+chbmv.mkl : chbmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+chbmv.veclib : chbmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+##################################### Zhbmv ####################################################
+
+zhbmv.goto : zhbmv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+zhbmv.acml : zhbmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zhbmv.atlas : zhbmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zhbmv.mkl : zhbmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zhbmv.veclib : zhbmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+##################################### Chpmv ####################################################
+
+chpmv.goto : chpmv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+chpmv.acml : chpmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+chpmv.atlas : chpmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+chpmv.mkl : chpmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+chpmv.veclib : chpmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+##################################### Zhpmv ####################################################
+
+zhpmv.goto : zhpmv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+zhpmv.acml : zhpmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zhpmv.atlas : zhpmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zhpmv.mkl : zhpmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zhpmv.veclib : zhpmv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+##################################### Sdot ####################################################
+sdot.goto : sdot.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+sdot.acml : sdot.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sdot.atlas : sdot.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sdot.mkl : sdot.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sdot.veclib : sdot.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Ddot ####################################################
+ddot.goto : ddot.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+ddot.acml : ddot.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ddot.atlas : ddot.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ddot.mkl : ddot.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ddot.veclib : ddot.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Cdot ####################################################
+cdot.goto : cdot.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+cdot.acml : cdot.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cdot.atlas : cdot.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cdot.mkl : cdot-intel.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cdot.veclib : cdot-intel.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Zdot ####################################################
+zdot.goto : zdot.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+zdot.acml : zdot.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zdot.atlas : zdot.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zdot.mkl : zdot-intel.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zdot.veclib : zdot-intel.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Srot ####################################################
+srot.goto : srot.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+srot.acml : srot.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+srot.atlas : srot.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+srot.mkl : srot.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+srot.veclib : srot.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Drot ####################################################
+drot.goto : drot.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+drot.acml : drot.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+drot.atlas : drot.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+drot.mkl : drot.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+drot.veclib : drot.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### csrot ####################################################
+csrot.goto : csrot.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+csrot.acml : csrot.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+csrot.atlas : csrot.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+csrot.mkl : csrot.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+csrot.veclib : csrot.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### zdrot ####################################################
+zdrot.goto : zdrot.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+zdrot.acml : zdrot.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zdrot.atlas : zdrot.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zdrot.mkl : zdrot.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zdrot.veclib : zdrot.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+	
+##################################### srotm ####################################################
+srotm.goto : srotm.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+srotm.acml : srotm.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+srotm.atlas : srotm.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+srotm.mkl : srotm.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+srotm.veclib : srotm.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### drotm ####################################################
+drotm.goto : drotm.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+drotm.acml : drotm.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+drotm.atlas : drotm.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+drotm.mkl : drotm.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+drotm.veclib : drotm.$(SUFFIX)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Saxpy ####################################################
+saxpy.goto : saxpy.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+saxpy.acml : saxpy.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+saxpy.atlas : saxpy.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+saxpy.mkl : saxpy.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+saxpy.veclib : saxpy.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Daxpy ####################################################
+daxpy.goto : daxpy.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+daxpy.acml : daxpy.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+daxpy.atlas : daxpy.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+daxpy.mkl : daxpy.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+daxpy.veclib : daxpy.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Caxpy ####################################################
+
+caxpy.goto : caxpy.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+caxpy.acml : caxpy.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+caxpy.atlas : caxpy.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+caxpy.mkl : caxpy.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+caxpy.veclib : caxpy.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Zaxpy ####################################################
+
+zaxpy.goto : zaxpy.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+zaxpy.acml : zaxpy.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zaxpy.atlas : zaxpy.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zaxpy.mkl : zaxpy.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zaxpy.veclib : zaxpy.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Saxpby ####################################################
+saxpby.goto : saxpby.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+saxpby.acml : saxpby.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+saxpby.atlas : saxpby.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+saxpby.mkl : saxpby.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+saxpby.veclib : saxpby.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Daxpby ####################################################
+daxpby.goto : daxpby.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+daxpby.acml : daxpby.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+daxpby.atlas : daxpby.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+daxpby.mkl : daxpby.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+daxpby.veclib : daxpby.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Caxpby ####################################################
+
+caxpby.goto : caxpby.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+caxpby.acml : caxpby.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+caxpby.atlas : caxpby.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+caxpby.mkl : caxpby.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+caxpby.veclib : caxpby.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Zaxpby ####################################################
+
+zaxpby.goto : zaxpby.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+zaxpby.acml : zaxpby.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zaxpby.atlas : zaxpby.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zaxpby.mkl : zaxpby.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zaxpby.veclib : zaxpby.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+	
+##################################### Scopy ####################################################
+scopy.goto : scopy.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+scopy.acml : scopy.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+scopy.atlas : scopy.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+scopy.mkl : scopy.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+scopy.veclib : scopy.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Dcopy ####################################################
+dcopy.goto : dcopy.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+dcopy.acml : dcopy.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dcopy.atlas : dcopy.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dcopy.mkl : dcopy.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dcopy.veclib : dcopy.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Ccopy ####################################################
+
+ccopy.goto : ccopy.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+ccopy.acml : ccopy.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ccopy.atlas : ccopy.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ccopy.mkl : ccopy.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+ccopy.veclib : ccopy.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Zcopy ####################################################
+
+zcopy.goto : zcopy.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+zcopy.acml : zcopy.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zcopy.atlas : zcopy.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zcopy.mkl : zcopy.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zcopy.veclib : zcopy.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Sscal ####################################################
+sscal.goto : sscal.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+sscal.acml : sscal.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sscal.atlas : sscal.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sscal.mkl : sscal.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sscal.veclib : sscal.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Dscal ####################################################
+dscal.goto : dscal.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+dscal.acml : dscal.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dscal.atlas : dscal.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dscal.mkl : dscal.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dscal.veclib : dscal.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Cscal ####################################################
+
+cscal.goto : cscal.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+cscal.acml : cscal.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cscal.atlas : cscal.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cscal.mkl : cscal.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cscal.veclib : cscal.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Zscal ####################################################
+
+zscal.goto : zscal.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+zscal.acml : zscal.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zscal.atlas : zscal.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zscal.mkl : zscal.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zscal.veclib : zscal.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Sasum ####################################################
+sasum.goto : sasum.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+sasum.acml : sasum.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sasum.atlas : sasum.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sasum.mkl : sasum.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sasum.veclib : sasum.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Dasum ####################################################
+dasum.goto : dasum.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+dasum.acml : dasum.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dasum.atlas : dasum.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dasum.mkl : dasum.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dasum.veclib : dasum.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Casum ####################################################
+
+casum.goto : casum.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+casum.acml : casum.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+casum.atlas : casum.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+casum.mkl : casum.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+casum.veclib : casum.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Zasum ####################################################
+
+zasum.goto : zasum.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+zasum.acml : zasum.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zasum.atlas : zasum.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zasum.mkl : zasum.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zasum.veclib : zasum.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Sswap ####################################################
+sswap.goto : sswap.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+sswap.acml : sswap.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sswap.atlas : sswap.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sswap.mkl : sswap.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sswap.veclib : sswap.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Dswap ####################################################
+dswap.goto : dswap.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+dswap.acml : dswap.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dswap.atlas : dswap.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dswap.mkl : dswap.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dswap.veclib : dswap.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Cswap ####################################################
+
+cswap.goto : cswap.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+cswap.acml : cswap.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cswap.atlas : cswap.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cswap.mkl : cswap.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cswap.veclib : cswap.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Zswap ####################################################
+
+zswap.goto : zswap.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+zswap.acml : zswap.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zswap.atlas : zswap.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zswap.mkl : zswap.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zswap.veclib : zswap.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+
+##################################### Sgesv ####################################################
+sgesv.goto : sgesv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+sgesv.acml : sgesv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sgesv.atlas : sgesv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sgesv.mkl : sgesv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+sgesv.veclib : sgesv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Dgesv ####################################################
+dgesv.goto : dgesv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+dgesv.acml : dgesv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dgesv.atlas : dgesv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dgesv.mkl : dgesv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+dgesv.veclib : dgesv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Cgesv ####################################################
+
+cgesv.goto : cgesv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+cgesv.acml : cgesv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cgesv.atlas : cgesv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cgesv.mkl : cgesv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cgesv.veclib : cgesv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Zgesv ####################################################
+
+zgesv.goto : zgesv.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+zgesv.acml : zgesv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zgesv.atlas : zgesv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zgesv.mkl : zgesv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zgesv.veclib : zgesv.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+
+##################################### Cgemm3m ####################################################
+
+cgemm3m.goto : cgemm3m.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+cgemm3m.mkl : cgemm3m.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+cgemm3m.veclib : cgemm3m.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+##################################### Zgemm3m ####################################################
+
+zgemm3m.goto : zgemm3m.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+zgemm3m.mkl : zgemm3m.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+zgemm3m.veclib : zgemm3m.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+############################################## ISAMAX ##############################################
+isamax.goto : isamax.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+isamax.atlas : isamax.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+############################################## IDAMAX ##############################################
+idamax.goto : idamax.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+idamax.atlas : idamax.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+############################################## ICAMAX ##############################################
+icamax.goto : icamax.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+icamax.atlas : icamax.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+############################################## IZAMAX ##############################################
+izamax.goto : izamax.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+izamax.atlas : izamax.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+############################################## ISMAX ##############################################
+ismax.goto : ismax.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+############################################## IDMAX ##############################################
+idmax.goto : idmax.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+	
+############################################## ISAMIN ##############################################
+isamin.goto : isamin.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+############################################## IDAMIN ##############################################
+idamin.goto : idamin.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+############################################## ICAMIN ##############################################
+icamin.goto : icamin.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+############################################## IZAMIN ##############################################
+izamin.goto : izamin.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+############################################## ISMIN ##############################################
+ismin.goto : ismin.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+############################################## IDMIN ##############################################
+idmin.goto : idmin.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+############################################## SAMAX ##############################################
+samax.goto : samax.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+############################################## DAMAX ##############################################
+damax.goto : damax.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+############################################## CAMAX ##############################################
+camax.goto : camax.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+############################################## ZAMAX ##############################################
+zamax.goto : zamax.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+############################################## SMAX ##############################################
+smax.goto : smax.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+############################################## DMAX ##############################################
+dmax.goto : dmax.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+############################################## SAMIN ##############################################
+samin.goto : samin.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+############################################## DAMIN ##############################################
+damin.goto : damin.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+############################################## CAMIN ##############################################
+camin.goto : camin.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+############################################## ZAMIN ##############################################
+zamin.goto : zamin.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+############################################## SMIN ##############################################
+smin.goto : smin.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+############################################## DMIN ##############################################
+dmin.goto : dmin.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+############################################## SNRM2 ##############################################
+snrm2.goto : snrm2.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+snrm2.atlas : snrm2.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+############################################## DNRM2 ##############################################
+dnrm2.goto : dnrm2.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+dnrm2.atlas : dnrm2.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+############################################## Sscnrm2 ##############################################
+scnrm2.goto : scnrm2.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+scnrm2.atlas : scnrm2.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+############################################## Ddznrm2 ##############################################
+dznrm2.goto : dznrm2.$(SUFFIX) ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
+
+dznrm2.atlas : dznrm2.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
+
+###################################################################################################
+
+slinpack.$(SUFFIX) : linpack.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+dlinpack.$(SUFFIX) : linpack.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+clinpack.$(SUFFIX) : linpack.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+zlinpack.$(SUFFIX) : linpack.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+scholesky.$(SUFFIX) : cholesky.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+dcholesky.$(SUFFIX) : cholesky.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+ccholesky.$(SUFFIX) : cholesky.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+zcholesky.$(SUFFIX) : cholesky.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+ifeq ($(BUILD_HALF),1)
+shgemm.$(SUFFIX) : gemm.c
+	$(CC) $(CFLAGS) -c -DHALF -UCOMPLEX -UDOUBLE -o $(@F) $^
+endif
+
+sgemm.$(SUFFIX) : gemm.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+dgemm.$(SUFFIX) : gemm.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+cgemm.$(SUFFIX) : gemm.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+zgemm.$(SUFFIX) : gemm.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+ssymm.$(SUFFIX) : symm.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+dsymm.$(SUFFIX) : symm.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+csymm.$(SUFFIX) : symm.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+zsymm.$(SUFFIX) : symm.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+strmm.$(SUFFIX) : trmm.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+dtrmm.$(SUFFIX) : trmm.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+ctrmm.$(SUFFIX) : trmm.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+ztrmm.$(SUFFIX) : trmm.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+strsm.$(SUFFIX) : trsm.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+dtrsm.$(SUFFIX) : trsm.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+ctrsm.$(SUFFIX) : trsm.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+ztrsm.$(SUFFIX) : trsm.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+ssyr.$(SUFFIX) : syr.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+dsyr.$(SUFFIX) : syr.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+	
+sspr.$(SUFFIX) : spr.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+dspr.$(SUFFIX) : spr.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+	
+sspr2.$(SUFFIX) : spr2.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+dspr2.$(SUFFIX) : spr2.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+ssyr2.$(SUFFIX) : syr2.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+dsyr2.$(SUFFIX) : syr2.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+ssyrk.$(SUFFIX) : syrk.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+dsyrk.$(SUFFIX) : syrk.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+csyrk.$(SUFFIX) : syrk.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+zsyrk.$(SUFFIX) : syrk.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+ssyr2k.$(SUFFIX) : syr2k.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+dsyr2k.$(SUFFIX) : syr2k.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+csyr2k.$(SUFFIX) : syr2k.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+zsyr2k.$(SUFFIX) : syr2k.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+chemm.$(SUFFIX) : hemm.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+zhemm.$(SUFFIX) : hemm.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+cherk.$(SUFFIX) : herk.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+zherk.$(SUFFIX) : herk.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+cher2k.$(SUFFIX) : her2k.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+zher2k.$(SUFFIX) : her2k.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+cher.$(SUFFIX) : her.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+zher.$(SUFFIX) : her.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+cher2.$(SUFFIX) : her2.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+zher2.$(SUFFIX) : her2.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+sgemv.$(SUFFIX) : gemv.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+dgemv.$(SUFFIX) : gemv.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+cgemv.$(SUFFIX) : gemv.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+zgemv.$(SUFFIX) : gemv.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+sspmv.$(SUFFIX) : spmv.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+dspmv.$(SUFFIX) : spmv.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+strmv.$(SUFFIX) : trmv.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+dtrmv.$(SUFFIX) : trmv.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+ctrmv.$(SUFFIX) : trmv.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+ztrmv.$(SUFFIX) : trmv.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+stpmv.$(SUFFIX) : tpmv.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+dtpmv.$(SUFFIX) : tpmv.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+ctpmv.$(SUFFIX) : tpmv.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+ztpmv.$(SUFFIX) : tpmv.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+stpsv.$(SUFFIX) : tpsv.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+dtpsv.$(SUFFIX) : tpsv.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+ctpsv.$(SUFFIX) : tpsv.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+ztpsv.$(SUFFIX) : tpsv.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+strsv.$(SUFFIX) : trsv.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+dtrsv.$(SUFFIX) : trsv.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+ctrsv.$(SUFFIX) : trsv.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+ztrsv.$(SUFFIX) : trsv.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+sger.$(SUFFIX) : ger.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+dger.$(SUFFIX) : ger.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+cger.$(SUFFIX) : ger.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+zger.$(SUFFIX) : ger.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+
+ssymv.$(SUFFIX) : symv.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+dsymv.$(SUFFIX) : symv.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+csymv.$(SUFFIX) : symv.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+zsymv.$(SUFFIX) : symv.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+sgeev.$(SUFFIX) : geev.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+dgeev.$(SUFFIX) : geev.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+cgeev.$(SUFFIX) : geev.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+zgeev.$(SUFFIX) : geev.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+sgetri.$(SUFFIX) : getri.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+dgetri.$(SUFFIX) : getri.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+cgetri.$(SUFFIX) : getri.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+zgetri.$(SUFFIX) : getri.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+spotrf.$(SUFFIX) : potrf.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+dpotrf.$(SUFFIX) : potrf.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+cpotrf.$(SUFFIX) : potrf.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+zpotrf.$(SUFFIX) : potrf.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+chemv.$(SUFFIX) : hemv.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+zhemv.$(SUFFIX) : hemv.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+chbmv.$(SUFFIX) : hbmv.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+zhbmv.$(SUFFIX) : hbmv.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+chpmv.$(SUFFIX) : hpmv.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+zhpmv.$(SUFFIX) : hpmv.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+sdot.$(SUFFIX) : dot.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+ddot.$(SUFFIX) : dot.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+cdot.$(SUFFIX) : zdot.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+zdot.$(SUFFIX) : zdot.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+cdot-intel.$(SUFFIX) : zdot-intel.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+zdot-intel.$(SUFFIX) : zdot-intel.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+
+
+saxpy.$(SUFFIX) : axpy.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+daxpy.$(SUFFIX) : axpy.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+caxpy.$(SUFFIX) : axpy.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+zaxpy.$(SUFFIX) : axpy.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+saxpby.$(SUFFIX) : axpby.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+daxpby.$(SUFFIX) : axpby.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+caxpby.$(SUFFIX) : axpby.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+zaxpby.$(SUFFIX) : axpby.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+scopy.$(SUFFIX) : copy.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+dcopy.$(SUFFIX) : copy.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+ccopy.$(SUFFIX) : copy.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+zcopy.$(SUFFIX) : copy.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+sswap.$(SUFFIX) : swap.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+dswap.$(SUFFIX) : swap.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+cswap.$(SUFFIX) : swap.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+zswap.$(SUFFIX) : swap.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+
+
+sscal.$(SUFFIX) : scal.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+dscal.$(SUFFIX) : scal.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+cscal.$(SUFFIX) : scal.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+zscal.$(SUFFIX) : scal.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+sasum.$(SUFFIX) : asum.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+dasum.$(SUFFIX) : asum.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+casum.$(SUFFIX) : asum.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+zasum.$(SUFFIX) : asum.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+
+sgesv.$(SUFFIX) : gesv.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+dgesv.$(SUFFIX) : gesv.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+cgesv.$(SUFFIX) : gesv.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+zgesv.$(SUFFIX) : gesv.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+srot.$(SUFFIX) : rot.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+drot.$(SUFFIX) : rot.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+csrot.$(SUFFIX) : rot.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+zdrot.$(SUFFIX) : rot.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+	
+srotm.$(SUFFIX) : rotm.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+drotm.$(SUFFIX) : rotm.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+
+
+cgemm3m.$(SUFFIX) : gemm3m.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+zgemm3m.$(SUFFIX) : gemm3m.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+
+isamax.$(SUFFIX) : iamax.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+idamax.$(SUFFIX) : iamax.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+icamax.$(SUFFIX) : iamax.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+izamax.$(SUFFIX) : iamax.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+
+ismax.$(SUFFIX) : imax.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+idmax.$(SUFFIX) : imax.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+
+isamin.$(SUFFIX) : iamin.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+idamin.$(SUFFIX) : iamin.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+icamin.$(SUFFIX) : iamin.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+izamin.$(SUFFIX) : iamin.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+
+ismin.$(SUFFIX) : imin.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+idmin.$(SUFFIX) : imin.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+
+samax.$(SUFFIX) : amax.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX  -UDOUBLE -o $(@F) $^
+
+damax.$(SUFFIX) : amax.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX  -DDOUBLE -o $(@F) $^
+
+camax.$(SUFFIX) : amax.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX  -UDOUBLE -o $(@F) $^
+
+zamax.$(SUFFIX) : amax.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX  -DDOUBLE -o $(@F) $^
+
+
+smax.$(SUFFIX) : max.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX  -UDOUBLE -o $(@F) $^
+
+dmax.$(SUFFIX) : max.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX  -DDOUBLE -o $(@F) $^
+
+
+samin.$(SUFFIX) : amin.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+damin.$(SUFFIX) : amin.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+camin.$(SUFFIX) : amin.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+zamin.$(SUFFIX) : amin.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+
+smin.$(SUFFIX) : min.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+dmin.$(SUFFIX) : min.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+
+snrm2.$(SUFFIX) : nrm2.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
+
+dnrm2.$(SUFFIX) : nrm2.c
+	$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
+
+scnrm2.$(SUFFIX) : nrm2.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
+
+dznrm2.$(SUFFIX) : nrm2.c
+	$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
+
+
+smallscaling: smallscaling.c ../$(LIBNAME)
+	$(CC) $(CFLAGS) -o $(@F) $^ $(EXTRALIB) -fopenmp -lm -lpthread
+
+clean ::
+	@rm -f *.goto *.mkl *.acml *.atlas *.veclib *.essl smallscaling
+
+include $(TOPDIR)/Makefile.tail
diff --git a/benchmark/amax.c b/benchmark/amax.c
new file mode 100644
index 000000000..32f55ce83
--- /dev/null
+++ b/benchmark/amax.c
@@ -0,0 +1,191 @@
+/***************************************************************************
+Copyright (c) 2016, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include <stdio.h>
+#include <stdlib.h>
+#ifdef __CYGWIN32__
+#include <sys/time.h>
+#endif
+#include "common.h"
+
+
+#undef AMAX
+
+#ifdef COMPLEX
+#ifdef DOUBLE
+#define AMAX   BLASFUNC(dzamax)
+#else
+#define AMAX   BLASFUNC(scamax)
+#endif
+#else
+#ifdef DOUBLE
+#define AMAX   BLASFUNC(damax)
+#else
+#define AMAX   BLASFUNC(samax)
+#endif
+#endif
+
+#if defined(__WIN32__) || defined(__WIN64__)
+
+#ifndef DELTA_EPOCH_IN_MICROSECS
+#define DELTA_EPOCH_IN_MICROSECS 11644473600000000ULL
+#endif
+
+int gettimeofday(struct timeval *tv, void *tz){
+
+  FILETIME ft;
+  unsigned __int64 tmpres = 0;
+  static int tzflag;
+
+  if (NULL != tv)
+    {
+      GetSystemTimeAsFileTime(&ft);
+
+      tmpres |= ft.dwHighDateTime;
+      tmpres <<= 32;
+      tmpres |= ft.dwLowDateTime;
+
+      /*converting file time to unix epoch*/
+      tmpres /= 10;  /*convert into microseconds*/
+      tmpres -= DELTA_EPOCH_IN_MICROSECS;
+      tv->tv_sec = (long)(tmpres / 1000000UL);
+      tv->tv_usec = (long)(tmpres % 1000000UL);
+    }
+
+  return 0;
+}
+
+#endif
+
+#if !defined(__WIN32__) && !defined(__WIN64__) && !defined(__CYGWIN32__) && 0
+
+static void *huge_malloc(BLASLONG size){
+  int shmid;
+  void *address;
+
+#ifndef SHM_HUGETLB
+#define SHM_HUGETLB 04000
+#endif
+
+  if ((shmid =shmget(IPC_PRIVATE,
+		     (size + HUGE_PAGESIZE) & ~(HUGE_PAGESIZE - 1),
+		     SHM_HUGETLB | IPC_CREAT |0600)) < 0) {
+    printf( "Memory allocation failed(shmget).\n");
+    exit(1);
+  }
+
+  address = shmat(shmid, NULL, SHM_RND);
+
+  if ((BLASLONG)address == -1){
+    printf( "Memory allocation failed(shmat).\n");
+    exit(1);
+  }
+
+  shmctl(shmid, IPC_RMID, 0);
+
+  return address;
+}
+
+#define malloc huge_malloc
+
+#endif
+
+int main(int argc, char *argv[]){
+
+  FLOAT *x;
+  blasint m, i;
+  blasint inc_x=1;
+  int loops = 1;
+  int l;
+  char *p;
+
+
+  int from =   1;
+  int to   = 200;
+  int step =   1;
+
+  struct timeval start, stop;
+  double time1,timeg;
+
+  argc--;argv++;
+
+  if (argc > 0) { from     = atol(*argv);		argc--; argv++;}
+  if (argc > 0) { to       = MAX(atol(*argv), from);	argc--; argv++;}
+  if (argc > 0) { step     = atol(*argv);		argc--; argv++;}
+
+  if ((p = getenv("OPENBLAS_LOOPS")))  loops = atoi(p);
+  if ((p = getenv("OPENBLAS_INCX")))   inc_x = atoi(p);
+
+  fprintf(stderr, "From : %3d  To : %3d Step = %3d Inc_x = %d Loops = %d\n", from, to, step,inc_x,loops);
+
+  if (( x = (FLOAT *)malloc(sizeof(FLOAT) * to * abs(inc_x) * COMPSIZE)) == NULL){
+    fprintf(stderr,"Out of Memory!!\n");exit(1);
+  }
+
+#ifdef linux
+  srandom(getpid());
+#endif
+
+  fprintf(stderr, "   SIZE       Flops\n");
+
+  for(m = from; m <= to; m += step)
+  {
+
+   timeg=0;
+
+   fprintf(stderr, " %6d : ", (int)m);
+
+
+   for (l=0; l<loops; l++)
+   {
+
+   	for(i = 0; i < m * COMPSIZE * abs(inc_x); i++){
+			x[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+   	}
+
+    	gettimeofday( &start, (struct timezone *)0);
+    	AMAX (&m, x, &inc_x);
+    	gettimeofday( &stop, (struct timezone *)0);
+
+    	time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
+
+	timeg += time1;
+
+    }
+
+    timeg /= loops;
+
+    fprintf(stderr,
+	    " %10.2f MFlops %10.6f sec\n",
+	    COMPSIZE * sizeof(FLOAT) * 1. * (double)m / timeg * 1.e-6, timeg);
+
+  }
+
+  return 0;
+}
+
+// void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));
diff --git a/benchmark/amin.c b/benchmark/amin.c
new file mode 100644
index 000000000..218f0ea9f
--- /dev/null
+++ b/benchmark/amin.c
@@ -0,0 +1,192 @@
+/***************************************************************************
+Copyright (c) 2016, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include <stdio.h>
+#include <stdlib.h>
+#ifdef __CYGWIN32__
+#include <sys/time.h>
+#endif
+#include "common.h"
+
+
+#undef AMIN
+
+#ifdef COMPLEX
+#ifdef DOUBLE
+#define AMIN   BLASFUNC(dzamin)
+#else
+#define AMIN   BLASFUNC(scamin)
+#endif
+#else
+#ifdef DOUBLE
+#define AMIN   BLASFUNC(damin)
+#else
+#define AMIN   BLASFUNC(samin)
+#endif
+#endif
+
+#if defined(__WIN32__) || defined(__WIN64__)
+
+#ifndef DELTA_EPOCH_IN_MICROSECS
+#define DELTA_EPOCH_IN_MICROSECS 11644473600000000ULL
+#endif
+
+int gettimeofday(struct timeval *tv, void *tz){
+
+  FILETIME ft;
+  unsigned __int64 tmpres = 0;
+  static int tzflag;
+
+  if (NULL != tv)
+    {
+      GetSystemTimeAsFileTime(&ft);
+
+      tmpres |= ft.dwHighDateTime;
+      tmpres <<= 32;
+      tmpres |= ft.dwLowDateTime;
+
+      /*converting file time to unix epoch*/
+      tmpres /= 10;  /*convert into microseconds*/
+      tmpres -= DELTA_EPOCH_IN_MICROSECS;
+      tv->tv_sec = (long)(tmpres / 1000000UL);
+      tv->tv_usec = (long)(tmpres % 1000000UL);
+    }
+
+  return 0;
+}
+
+#endif
+
+#if !defined(__WIN32__) && !defined(__WIN64__) && !defined(__CYGWIN32__) && 0
+
+static void *huge_malloc(BLASLONG size){
+  int shmid;
+  void *address;
+
+#ifndef SHM_HUGETLB
+#define SHM_HUGETLB 04000
+#endif
+
+  if ((shmid =shmget(IPC_PRIVATE,
+		     (size + HUGE_PAGESIZE) & ~(HUGE_PAGESIZE - 1),
+		     SHM_HUGETLB | IPC_CREAT |0600)) < 0) {
+    printf( "Memory allocation failed(shmget).\n");
+    exit(1);
+  }
+
+  address = shmat(shmid, NULL, SHM_RND);
+
+  if ((BLASLONG)address == -1){
+    printf( "Memory allocation failed(shmat).\n");
+    exit(1);
+  }
+
+  shmctl(shmid, IPC_RMID, 0);
+
+  return address;
+}
+
+#define malloc huge_malloc
+
+#endif
+
+int main(int argc, char *argv[]){
+
+  FLOAT *x;
+  blasint m, i;
+  blasint inc_x=1;
+  int loops = 1;
+  int l;
+  char *p;
+
+  int from =   1;
+  int to   = 200;
+  int step =   1;
+
+  struct timeval start, stop;
+  double time1,timeg;
+
+  argc--;argv++;
+
+  if (argc > 0) { from     = atol(*argv);		argc--; argv++;}
+  if (argc > 0) { to       = MAX(atol(*argv), from);	argc--; argv++;}
+  if (argc > 0) { step     = atol(*argv);		argc--; argv++;}
+
+  if ((p = getenv("OPENBLAS_LOOPS")))  loops = atoi(p);
+  if ((p = getenv("OPENBLAS_INCX")))   inc_x = atoi(p);
+
+  fprintf(stderr, "From : %3d  To : %3d Step = %3d Inc_x = %d Loops = %d\n", from, to, step,inc_x,loops);
+
+  if (( x = (FLOAT *)malloc(sizeof(FLOAT) * to * abs(inc_x) * COMPSIZE)) == NULL){
+    fprintf(stderr,"Out of Memory!!\n");exit(1);
+  }
+
+#ifdef linux
+  srandom(getpid());
+#endif
+
+  fprintf(stderr, "   SIZE       Flops\n");
+
+  for(m = from; m <= to; m += step)
+  {
+
+   timeg=0;
+
+   fprintf(stderr, " %6d : ", (int)m);
+
+
+   for (l=0; l<loops; l++)
+   {
+
+   	for(i = 0; i < m * COMPSIZE * abs(inc_x); i++){
+			x[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+   	}
+
+    	gettimeofday( &start, (struct timezone *)0);
+
+    	AMIN (&m, x, &inc_x);
+
+    	gettimeofday( &stop, (struct timezone *)0);
+
+    	time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
+
+	timeg += time1;
+
+    }
+
+    timeg /= loops;
+
+    fprintf(stderr,
+	    " %10.2f MFlops %10.6f sec\n",
+	    COMPSIZE * sizeof(FLOAT) * 1. * (double)m / timeg * 1.e-6, timeg);
+
+  }
+
+  return 0;
+}
+
+// void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));
diff --git a/benchmark/axpby.c b/benchmark/axpby.c
new file mode 100644
index 000000000..3b3dd9979
--- /dev/null
+++ b/benchmark/axpby.c
@@ -0,0 +1,202 @@
+/***************************************************************************
+Copyright (c) 2014, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include <stdio.h>
+#include <stdlib.h>
+#ifdef __CYGWIN32__
+#include <sys/time.h>
+#endif
+#include "common.h"
+
+
+#undef AXPBY
+
+#ifdef COMPLEX
+#ifdef DOUBLE
+#define AXPBY   BLASFUNC(zaxpby)
+#else
+#define AXPBY   BLASFUNC(caxpby)
+#endif
+#else
+#ifdef DOUBLE
+#define AXPBY   BLASFUNC(daxpby)
+#else
+#define AXPBY   BLASFUNC(saxpby)
+#endif
+#endif
+
+#if defined(__WIN32__) || defined(__WIN64__)
+
+#ifndef DELTA_EPOCH_IN_MICROSECS
+#define DELTA_EPOCH_IN_MICROSECS 11644473600000000ULL
+#endif
+
+int gettimeofday(struct timeval *tv, void *tz){
+
+  FILETIME ft;
+  unsigned __int64 tmpres = 0;
+  static int tzflag;
+
+  if (NULL != tv)
+    {
+      GetSystemTimeAsFileTime(&ft);
+
+      tmpres |= ft.dwHighDateTime;
+      tmpres <<= 32;
+      tmpres |= ft.dwLowDateTime;
+
+      /*converting file time to unix epoch*/
+      tmpres /= 10;  /*convert into microseconds*/
+      tmpres -= DELTA_EPOCH_IN_MICROSECS;
+      tv->tv_sec = (long)(tmpres / 1000000UL);
+      tv->tv_usec = (long)(tmpres % 1000000UL);
+    }
+
+  return 0;
+}
+
+#endif
+
+#if !defined(__WIN32__) && !defined(__WIN64__) && !defined(__CYGWIN32__) && 0
+
+static void *huge_malloc(BLASLONG size){
+  int shmid;
+  void *address;
+
+#ifndef SHM_HUGETLB
+#define SHM_HUGETLB 04000
+#endif
+
+  if ((shmid =shmget(IPC_PRIVATE,
+             (size + HUGE_PAGESIZE) & ~(HUGE_PAGESIZE - 1),
+             SHM_HUGETLB | IPC_CREAT |0600)) < 0) {
+    printf( "Memory allocation failed(shmget).\n");
+    exit(1);
+  }
+
+  address = shmat(shmid, NULL, SHM_RND);
+
+  if ((BLASLONG)address == -1){
+    printf( "Memory allocation failed(shmat).\n");
+    exit(1);
+  }
+
+  shmctl(shmid, IPC_RMID, 0);
+
+  return address;
+}
+
+#define malloc huge_malloc
+
+#endif
+
+int main(int argc, char *argv[]){
+
+  FLOAT *x, *y;
+  FLOAT alpha[2] = { 2.0, 2.0 };
+  FLOAT beta[2] = {2.0, 2.0};
+  blasint m, i;
+  blasint inc_x=1,inc_y=1;
+  int loops = 1;
+  int l;
+  char *p;
+
+  int from =   1;
+  int to   = 200;
+  int step =   1;
+
+  struct timeval start, stop;
+  double time1,timeg;
+
+  argc--;argv++;
+
+  if (argc > 0) { from     = atol(*argv);       argc--; argv++;}
+  if (argc > 0) { to       = MAX(atol(*argv), from);    argc--; argv++;}
+  if (argc > 0) { step     = atol(*argv);       argc--; argv++;}
+
+  if ((p = getenv("OPENBLAS_LOOPS")))  loops = atoi(p);
+  if ((p = getenv("OPENBLAS_INCX")))   inc_x = atoi(p);
+  if ((p = getenv("OPENBLAS_INCY")))   inc_y = atoi(p);
+
+  fprintf(stderr, "From : %3d  To : %3d Step = %3d Inc_x = %d Inc_y = %d Loops = %d\n", from, to, step,inc_x,inc_y,loops);
+
+  if (( x = (FLOAT *)malloc(sizeof(FLOAT) * to * abs(inc_x) * COMPSIZE)) == NULL){
+    fprintf(stderr,"Out of Memory!!\n");exit(1);
+  }
+
+  if (( y = (FLOAT *)malloc(sizeof(FLOAT) * to * abs(inc_y) * COMPSIZE)) == NULL){
+    fprintf(stderr,"Out of Memory!!\n");exit(1);
+  }
+
+#ifdef linux
+  srandom(getpid());
+#endif
+
+  fprintf(stderr, "   SIZE       Flops\n");
+
+  for(m = from; m <= to; m += step)
+  {
+
+    timeg=0;
+
+    fprintf(stderr, " %6d : ", (int)m);
+
+
+    for(i = 0; i < m * COMPSIZE * abs(inc_x); i++){
+            x[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+    }
+
+    for(i = 0; i < m * COMPSIZE * abs(inc_y); i++){
+            y[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+    }
+
+    for (l=0; l<loops; l++)
+    {
+        gettimeofday( &start, (struct timezone *)0);
+
+        AXPBY (&m, alpha, x, &inc_x, beta, y, &inc_y );
+
+        gettimeofday( &stop, (struct timezone *)0);
+
+        time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
+
+        timeg += time1;
+
+    }
+
+    timeg /= loops;
+
+    fprintf(stderr,
+        " %10.2f MFlops %10.6f sec\n",
+        (COMPSIZE * COMPSIZE * 4. - COMPSIZE) * (double)m / timeg * 1.e-6, timeg);
+
+  }
+
+  return 0;
+}
+
+// void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));
diff --git a/benchmark/axpy.c b/benchmark/axpy.c
index 37c7aeb63..e40f93c70 100644
--- a/benchmark/axpy.c
+++ b/benchmark/axpy.c
@@ -128,7 +128,7 @@ int main(int argc, char *argv[]){
   int to   = 200;
   int step =   1;
 
-  struct timeval start, stop;
+  struct timespec start, stop;
   double time1,timeg;
 
   argc--;argv++;
@@ -175,13 +175,13 @@ int main(int argc, char *argv[]){
    	for(i = 0; i < m * COMPSIZE * abs(inc_y); i++){
 			y[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
    	}
-    	gettimeofday( &start, (struct timezone *)0);
+    	clock_gettime( CLOCK_REALTIME, &start);
 
     	AXPY (&m, alpha, x, &inc_x, y, &inc_y );
 
-    	gettimeofday( &stop, (struct timezone *)0);
+    	clock_gettime( CLOCK_REALTIME, &stop);
 
-    	time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
+    	time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_nsec - start.tv_nsec)) * 1.e-9;
 
 	timeg += time1;
 
@@ -190,7 +190,7 @@ int main(int argc, char *argv[]){
     timeg /= loops;
 
     fprintf(stderr,
-	    " %10.2f MFlops %10.6f sec\n",
+	    " %10.2f MFlops %10.9f sec\n",
 	    COMPSIZE * COMPSIZE * 2. * (double)m / timeg * 1.e-6, timeg);
 
   }
diff --git a/benchmark/cholesky.c b/benchmark/cholesky.c
index 8d121efb3..5908b6085 100644
--- a/benchmark/cholesky.c
+++ b/benchmark/cholesky.c
@@ -173,46 +173,46 @@ int main(int argc, char *argv[]){
 #ifndef COMPLEX
       if (uplos & 1) {
 	for (j = 0; j < m; j++) {
-	  for(i = 0; i < j; i++)     a[i + j * m] = 0.;
-	                             a[j + j * m] = ((double) rand() / (double) RAND_MAX) + 8.;
-	  for(i = j + 1; i < m; i++) a[i + j * m] = ((double) rand() / (double) RAND_MAX) - 0.5;
+	  for(i = 0; i < j; i++)     a[(long)i + (long)j * (long)m] = 0.;
+	                             a[(long)j + (long)j * (long)m] = ((double) rand() / (double) RAND_MAX) + 8.;
+	  for(i = j + 1; i < m; i++) a[(long)i + (long)j * (long)m] = ((double) rand() / (double) RAND_MAX) - 0.5;
 	}
       } else {
 	for (j = 0; j < m; j++) {
-	  for(i = 0; i < j; i++)     a[i + j * m] = ((double) rand() / (double) RAND_MAX) - 0.5;
-	                             a[j + j * m] = ((double) rand() / (double) RAND_MAX) + 8.;
-	  for(i = j + 1; i < m; i++) a[i + j * m] = 0.;
+	  for(i = 0; i < j; i++)     a[(long)i + (long)j * (long)m] = ((double) rand() / (double) RAND_MAX) - 0.5;
+	                             a[(long)j + (long)j * (long)m] = ((double) rand() / (double) RAND_MAX) + 8.;
+	  for(i = j + 1; i < m; i++) a[(long)i + (long)j * (long)m] = 0.;
 	}
       }
 #else
       if (uplos & 1) {
 	for (j = 0; j < m; j++) {
 	  for(i = 0; i < j; i++) {
-	    a[(i + j * m) * 2 + 0] = 0.;
-	    a[(i + j * m) * 2 + 1] = 0.;
+	    a[((long)i + (long)j * (long)m) * 2 + 0] = 0.;
+	    a[((long)i + (long)j * (long)m) * 2 + 1] = 0.;
 	  }
 
-	  a[(j + j * m) * 2 + 0] = ((double) rand() / (double) RAND_MAX) + 8.;
-	  a[(j + j * m) * 2 + 1] = 0.;
+	  a[((long)j + (long)j * (long)m) * 2 + 0] = ((double) rand() / (double) RAND_MAX) + 8.;
+	  a[((long)j + (long)j * (long)m) * 2 + 1] = 0.;
 
 	  for(i = j + 1; i < m; i++) {
-	    a[(i + j * m) * 2 + 0] = ((double) rand() / (double) RAND_MAX) - 0.5;
-	    a[(i + j * m) * 2 + 1] = ((double) rand() / (double) RAND_MAX) - 0.5;
+	    a[((long)i + (long)j * (long)m) * 2 + 0] = ((double) rand() / (double) RAND_MAX) - 0.5;
+	    a[((long)i + (long)j * (long)m) * 2 + 1] = ((double) rand() / (double) RAND_MAX) - 0.5;
 	  }
 	}
       } else {
 	for (j = 0; j < m; j++) {
 	  for(i = 0; i < j; i++) {
-	    a[(i + j * m) * 2 + 0] = ((double) rand() / (double) RAND_MAX) - 0.5;
-	    a[(i + j * m) * 2 + 1] = ((double) rand() / (double) RAND_MAX) - 0.5;
+	    a[((long)i + (long)j * (long)m) * 2 + 0] = ((double) rand() / (double) RAND_MAX) - 0.5;
+	    a[((long)i + (long)j * (long)m) * 2 + 1] = ((double) rand() / (double) RAND_MAX) - 0.5;
 	  }
 
-	  a[(j + j * m) * 2 + 0] = ((double) rand() / (double) RAND_MAX) + 8.;
-	  a[(j + j * m) * 2 + 1] = 0.;
+	  a[((long)j + (long)j * (long)m) * 2 + 0] = ((double) rand() / (double) RAND_MAX) + 8.;
+	  a[((long)j + (long)j * (long)m) * 2 + 1] = 0.;
 
 	  for(i = j + 1; i < m; i++) {
-	    a[(i + j * m) * 2 + 0] = 0.;
-	    a[(i + j * m) * 2 + 1] = 0.;
+	    a[((long)i + (long)j * (long)m) * 2 + 0] = 0.;
+	    a[((long)i + (long)j * (long)m) * 2 + 1] = 0.;
 	  }
 	}
       }
@@ -239,10 +239,13 @@ int main(int argc, char *argv[]){
 	for (j = 0; j < m; j++) {
 	  for(i = 0; i <= j; i++) {
 #ifndef COMPLEX
-	    if (maxerr < fabs(a[i + j * m] - b[i + j * m])) maxerr = fabs(a[i + j * m] - b[i + j * m]);
+	    if (maxerr < fabs(a[(long)i + (long)j * (long)m] - b[(long)i + (long)j * (long)m]))
+	        maxerr = fabs(a[(long)i + (long)j * (long)m] - b[(long)i + (long)j * (long)m]);
 #else
-	    if (maxerr < fabs(a[(i + j * m) * 2 + 0] - b[(i + j * m) * 2 + 0])) maxerr = fabs(a[(i + j * m) * 2 + 0] - b[(i + j * m) * 2 + 0]);
-	    if (maxerr < fabs(a[(i + j * m) * 2 + 1] - b[(i + j * m) * 2 + 1])) maxerr = fabs(a[(i + j * m) * 2 + 1] - b[(i + j * m) * 2 + 1]);
+	    if (maxerr < fabs(a[((long)i + (long)j * (long)m) * 2 + 0] - b[((long)i + (long)j * (long)m) * 2 + 0]))
+	        maxerr = fabs(a[((long)i + (long)j * (long)m) * 2 + 0] - b[((long)i + (long)j * (long)m) * 2 + 0]);
+	    if (maxerr < fabs(a[((long)i + (long)j * (long)m) * 2 + 1] - b[((long)i + (long)j * (long)m) * 2 + 1]))
+	        maxerr = fabs(a[((long)i + (long)j * (long)m) * 2 + 1] - b[((long)i + (long)j * (long)m) * 2 + 1]);
 #endif
 	  }
 	}
@@ -250,10 +253,13 @@ int main(int argc, char *argv[]){
 	for (j = 0; j < m; j++) {
 	  for(i = j; i < m; i++) {
 #ifndef COMPLEX
-	    if (maxerr < fabs(a[i + j * m] - b[i + j * m])) maxerr = fabs(a[i + j * m] - b[i + j * m]);
+	    if (maxerr < fabs(a[(long)i + (long)j * (long)m] - b[(long)i + (long)j * (long)m]))
+	        maxerr = fabs(a[(long)i + (long)j * (long)m] - b[(long)i + (long)j * (long)m]);
 #else
-	    if (maxerr < fabs(a[(i + j * m) * 2 + 0] - b[(i + j * m) * 2 + 0])) maxerr = fabs(a[(i + j * m) * 2 + 0] - b[(i + j * m) * 2 + 0]);
-	    if (maxerr < fabs(a[(i + j * m) * 2 + 1] - b[(i + j * m) * 2 + 1])) maxerr = fabs(a[(i + j * m) * 2 + 1] - b[(i + j * m) * 2 + 1]);
+	    if (maxerr < fabs(a[((long)i + (long)j * (long)m) * 2 + 0] - b[((long)i + (long)j * (long)m) * 2 + 0]))
+	        maxerr = fabs(a[((long)i + (long)j * (long)m) * 2 + 0] - b[((long)i + (long)j * (long)m) * 2 + 0]);
+	    if (maxerr < fabs(a[((long)i + (long)j * (long)m) * 2 + 1] - b[((long)i + (long)j * (long)m) * 2 + 1]))
+	        maxerr = fabs(a[((long)i + (long)j * (long)m) * 2 + 1] - b[((long)i + (long)j * (long)m) * 2 + 1]);
 #endif
 	  }
 	}
diff --git a/benchmark/geev.c b/benchmark/geev.c
index d3751defb..ef9271220 100644
--- a/benchmark/geev.c
+++ b/benchmark/geev.c
@@ -195,7 +195,7 @@ int main(int argc, char *argv[]){
 
     for(j = 0; j < to; j++){
       for(i = 0; i < to * COMPSIZE; i++){
-	a[i + j * to * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+	a[(long)i + (long)j * (long)to * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
       }
     }
 
diff --git a/benchmark/gemm.c b/benchmark/gemm.c
index dd016a7c3..d2235330b 100644
--- a/benchmark/gemm.c
+++ b/benchmark/gemm.c
@@ -39,6 +39,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #ifdef DOUBLE
 #define GEMM   BLASFUNC(dgemm)
+#elif defined(HALF)
+#define GEMM   BLASFUNC(shgemm)
 #else
 #define GEMM   BLASFUNC(sgemm)
 #endif
@@ -120,7 +122,8 @@ static void *huge_malloc(BLASLONG size){
 
 int main(int argc, char *argv[]){
 
-  FLOAT *a, *b, *c;
+  IFLOAT *a, *b;
+  FLOAT *c;
   FLOAT alpha[] = {1.0, 0.0};
   FLOAT beta [] = {0.0, 0.0};
   char transa = 'N';
@@ -184,10 +187,10 @@ int main(int argc, char *argv[]){
     k = to;
   }
 
-  if (( a = (FLOAT *)malloc(sizeof(FLOAT) * m * k * COMPSIZE)) == NULL) {
+  if (( a = (IFLOAT *)malloc(sizeof(IFLOAT) * m * k * COMPSIZE)) == NULL) {
     fprintf(stderr,"Out of Memory!!\n");exit(1);
   }
-  if (( b = (FLOAT *)malloc(sizeof(FLOAT) * k * n * COMPSIZE)) == NULL) {
+  if (( b = (IFLOAT *)malloc(sizeof(IFLOAT) * k * n * COMPSIZE)) == NULL) {
     fprintf(stderr,"Out of Memory!!\n");exit(1);
   }
   if (( c = (FLOAT *)malloc(sizeof(FLOAT) * m * n * COMPSIZE)) == NULL) {
@@ -199,10 +202,10 @@ int main(int argc, char *argv[]){
 #endif
 
   for (i = 0; i < m * k * COMPSIZE; i++) {
-    a[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+    a[i] = ((IFLOAT) rand() / (IFLOAT) RAND_MAX) - 0.5;
   }
   for (i = 0; i < k * n * COMPSIZE; i++) {
-    b[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+    b[i] = ((IFLOAT) rand() / (IFLOAT) RAND_MAX) - 0.5;
   }
   for (i = 0; i < m * n * COMPSIZE; i++) {
     c[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
diff --git a/benchmark/gemm3m.c b/benchmark/gemm3m.c
index d39543585..f4048c436 100644
--- a/benchmark/gemm3m.c
+++ b/benchmark/gemm3m.c
@@ -181,9 +181,9 @@ int main(int argc, char *argv[]){
   
     	for(j = 0; j < m; j++){
       		for(i = 0; i < m * COMPSIZE; i++){
-			a[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
-			b[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
-			c[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+			a[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+			b[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+			c[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
       		}
     	}
 
diff --git a/benchmark/gemv.c b/benchmark/gemv.c
index adf8f3d91..a9dee67d2 100644
--- a/benchmark/gemv.c
+++ b/benchmark/gemv.c
@@ -197,7 +197,7 @@ int main(int argc, char *argv[]){
    		fprintf(stderr, " %6dx%d : ", (int)m,(int)n);
    		for(j = 0; j < m; j++){
       			for(i = 0; i < n * COMPSIZE; i++){
-				a[j + i * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+				a[(long)j + (long)i * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
       			}
    		}
 
@@ -234,7 +234,7 @@ int main(int argc, char *argv[]){
    		fprintf(stderr, " %6dx%d : ", (int)m,(int)n);
    		for(j = 0; j < m; j++){
       			for(i = 0; i < n * COMPSIZE; i++){
-				a[j + i * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+				a[(long)j + (long)i * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
       			}
    		}
 
diff --git a/benchmark/ger.c b/benchmark/ger.c
index a752a3c3e..ca7e94e15 100644
--- a/benchmark/ger.c
+++ b/benchmark/ger.c
@@ -182,7 +182,7 @@ int main(int argc, char *argv[]){
 
    for(j = 0; j < m; j++){
       		for(i = 0; i < n * COMPSIZE; i++){
-			a[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+			a[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
       		}
    }
 
diff --git a/benchmark/gesv.c b/benchmark/gesv.c
index 26ff8bc1a..80f644e69 100644
--- a/benchmark/gesv.c
+++ b/benchmark/gesv.c
@@ -177,20 +177,20 @@ int main(int argc, char *argv[]){
 
     for(j = 0; j < m; j++){
       for(i = 0; i < m * COMPSIZE; i++){
-	a[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+	a[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
       }
     }
 
     for(j = 0; j < m; j++){
       for(i = 0; i < m * COMPSIZE; i++){
-	b[i + j * m * COMPSIZE] = 0.0;
+	b[(long)i + (long)j * (long)m * COMPSIZE] = 0.0;
       }
     }
 
 
     for (j = 0; j < m; ++j) {
       for (i = 0; i < m * COMPSIZE; ++i) {
-	b[i] += a[i + j * m * COMPSIZE];
+	b[i] += a[(long)i + (long)j * (long)m * COMPSIZE];
       }
     }
 
diff --git a/benchmark/getri.c b/benchmark/getri.c
index 083cdc9aa..e8b82a758 100644
--- a/benchmark/getri.c
+++ b/benchmark/getri.c
@@ -172,7 +172,7 @@ int main(int argc, char *argv[]){
 
     for(j = 0; j < to; j++){
       for(i = 0; i < to * COMPSIZE; i++){
-	a[i + j * to * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+	a[(long)i + (long)j * (long)to * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
       }
     }
 
diff --git a/benchmark/hbmv.c b/benchmark/hbmv.c
new file mode 100644
index 000000000..b9dcc03bb
--- /dev/null
+++ b/benchmark/hbmv.c
@@ -0,0 +1,210 @@
+/***************************************************************************
+Copyright (c) 2014, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include <stdio.h>
+#include <stdlib.h>
+#ifdef __CYGWIN32__
+#include <sys/time.h>
+#endif
+#include "common.h"
+
+
+#undef HBMV
+
+
+#ifdef DOUBLE
+#define HBMV   BLASFUNC(zhbmv)
+#else
+#define HBMV   BLASFUNC(chbmv)
+#endif
+
+
+#if defined(__WIN32__) || defined(__WIN64__)
+
+#ifndef DELTA_EPOCH_IN_MICROSECS
+#define DELTA_EPOCH_IN_MICROSECS 11644473600000000ULL
+#endif
+
+int gettimeofday(struct timeval *tv, void *tz) {
+
+    FILETIME ft;
+    unsigned __int64 tmpres = 0;
+    static int tzflag;
+
+    if (NULL != tv)
+    {
+        GetSystemTimeAsFileTime(&ft);
+
+        tmpres |= ft.dwHighDateTime;
+        tmpres <<= 32;
+        tmpres |= ft.dwLowDateTime;
+
+        /*converting file time to unix epoch*/
+        tmpres /= 10;  /*convert into microseconds*/
+        tmpres -= DELTA_EPOCH_IN_MICROSECS;
+        tv->tv_sec = (long)(tmpres / 1000000UL);
+        tv->tv_usec = (long)(tmpres % 1000000UL);
+    }
+
+    return 0;
+}
+
+#endif
+
+#if !defined(__WIN32__) && !defined(__WIN64__) && !defined(__CYGWIN32__) && 0
+
+static void *huge_malloc(BLASLONG size) {
+    int shmid;
+    void *address;
+
+#ifndef SHM_HUGETLB
+#define SHM_HUGETLB 04000
+#endif
+
+    if ((shmid =shmget(IPC_PRIVATE,
+                (size + HUGE_PAGESIZE) & ~(HUGE_PAGESIZE - 1),
+                SHM_HUGETLB | IPC_CREAT |0600)) < 0) {
+        printf( "Memory allocation failed(shmget).\n");
+        exit(1);
+    }
+
+    address = shmat(shmid, NULL, SHM_RND);
+
+    if ((BLASLONG)address == -1){
+        printf( "Memory allocation failed(shmat).\n");
+        exit(1);
+    }
+
+    shmctl(shmid, IPC_RMID, 0);
+
+    return address;
+}
+
+#define malloc huge_malloc
+
+#endif
+
+int main(int argc, char *argv[]){
+
+    FLOAT *a, *x, *y;
+    FLOAT alpha[] = {1.0, 1.0};
+    FLOAT beta [] = {0.0, 0.0};
+    blasint k = 1;
+    char uplo='L';
+    blasint m, i, j;
+    blasint inc_x=1, inc_y=1;
+    int loops = 1;
+    int l;
+    char *p;
+
+    int from =   1;
+    int to   = 200;
+    int step =   1;
+
+    struct timeval start, stop;
+    double time1,timeg;
+
+    argc--;argv++;
+
+    if (argc > 0) { from     = atol(*argv);		argc--; argv++;}
+    if (argc > 0) { to       = MAX(atol(*argv), from);	argc--; argv++;}
+    if (argc > 0) { step     = atol(*argv);		argc--; argv++;}
+
+    if ((p = getenv("OPENBLAS_LOOPS")))  loops = atoi(p);
+    if ((p = getenv("OPENBLAS_INCX")))   inc_x = atoi(p);
+    if ((p = getenv("OPENBLAS_INCY")))   inc_y = atoi(p);
+    if ((p = getenv("OPENBLAS_UPLO")))   uplo=*p;
+    if ((p = getenv("OPENBLAS_K")))      k = atoi(p);
+    
+    fprintf(stderr, "From : %3d  To : %3d Step = %3d Uplo = '%c' k = %d Inc_x = %d Inc_y = %d Loops = %d\n", 
+                            from, to, step, uplo, k, inc_x, inc_y, loops);
+
+    if (( a = (FLOAT *)malloc(sizeof(FLOAT) * to * to * COMPSIZE)) == NULL) {
+        fprintf(stderr,"Out of Memory!!\n");
+        exit(1);
+    }
+
+    if (( x = (FLOAT *)malloc(sizeof(FLOAT) * to * abs(inc_x) * COMPSIZE)) == NULL) {
+        fprintf(stderr,"Out of Memory!!\n");
+        exit(1);
+    }
+
+    if (( y = (FLOAT *)malloc(sizeof(FLOAT) * to * abs(inc_y) * COMPSIZE)) == NULL) {
+        fprintf(stderr,"Out of Memory!!\n");
+        exit(1);
+    }
+
+#ifdef linux
+    srandom(getpid());
+#endif
+
+    fprintf(stderr, "   SIZE       Flops\n");
+
+    for(m = from; m <= to; m += step) {
+
+        timeg=0;
+
+        fprintf(stderr, " %6dx%d : ", (int)m, (int)m);
+
+        for(j = 0; j < m; j++) {
+            for(i = 0; i < m * COMPSIZE; i++) {
+                a[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+            }
+        }
+
+    for (l = 0; l < loops; l++) {
+
+        for (i = 0; i < m * COMPSIZE * abs(inc_x); i++) {
+            x[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+        }
+
+        for (i = 0; i < m * COMPSIZE * abs(inc_y); i++) {
+            y[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+        }
+
+        gettimeofday( &start, (struct timezone *)0);
+
+        HBMV (&uplo, &m, &k, alpha, a, &m, x, &inc_x, beta, y, &inc_y );
+
+        gettimeofday( &stop, (struct timezone *)0);
+
+        time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
+
+        timeg += time1;
+
+    }
+
+    timeg /= loops;
+
+    fprintf(stderr, " %10.2f MFlops\n",
+            COMPSIZE * COMPSIZE * 2. * (double)(2 * k + 1) * (double)m / timeg * 1.e-6);
+    }
+
+    return 0;
+}
+
+// void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));
diff --git a/benchmark/hemm.c b/benchmark/hemm.c
index 318c407ba..2fe0f5c5f 100644
--- a/benchmark/hemm.c
+++ b/benchmark/hemm.c
@@ -164,9 +164,9 @@ int main(int argc, char *argv[]){
 
     for(j = 0; j < m; j++){
       for(i = 0; i < m * COMPSIZE; i++){
-	a[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
-	b[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
-	c[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+	a[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+	b[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+	c[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
       }
     }
 
@@ -178,8 +178,6 @@ int main(int argc, char *argv[]){
 
     time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
 
-    gettimeofday( &start, (struct timezone *)0);
-
     fprintf(stderr,
 	    " %10.2f MFlops\n",
 	    COMPSIZE * COMPSIZE * 2. * (double)m * (double)m * (double)m / time1 * 1.e-6);
diff --git a/benchmark/hemv.c b/benchmark/hemv.c
index 05028e3cf..b6ff512ce 100644
--- a/benchmark/hemv.c
+++ b/benchmark/hemv.c
@@ -167,7 +167,7 @@ int main(int argc, char *argv[]){
 
    for(j = 0; j < m; j++){
       		for(i = 0; i < m * COMPSIZE; i++){
-			a[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+			a[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
       		}
    }
 
diff --git a/benchmark/her.c b/benchmark/her.c
new file mode 100644
index 000000000..f4e10b684
--- /dev/null
+++ b/benchmark/her.c
@@ -0,0 +1,186 @@
+/***************************************************************************
+Copyright (c) 2020, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include <stdio.h>
+#include <stdlib.h>
+#ifdef __CYGWIN32__
+#include <sys/time.h>
+#endif
+#include "common.h"
+
+
+#undef HER
+
+
+#ifdef DOUBLE
+#define HER   BLASFUNC(zher)
+#else
+#define HER   BLASFUNC(cher)
+#endif
+
+
+#if defined(__WIN32__) || defined(__WIN64__)
+
+#ifndef DELTA_EPOCH_IN_MICROSECS
+#define DELTA_EPOCH_IN_MICROSECS 11644473600000000ULL
+#endif
+
+int gettimeofday(struct timeval *tv, void *tz){
+
+    FILETIME ft;
+    unsigned __int64 tmpres = 0;
+    static int tzflag;
+
+    if (NULL != tv)
+    {
+        GetSystemTimeAsFileTime(&ft);
+
+        tmpres |= ft.dwHighDateTime;
+        tmpres <<= 32;
+        tmpres |= ft.dwLowDateTime;
+
+        /*converting file time to unix epoch*/
+        tmpres /= 10;  /*convert into microseconds*/
+        tmpres -= DELTA_EPOCH_IN_MICROSECS;
+        tv->tv_sec = (long)(tmpres / 1000000UL);
+        tv->tv_usec = (long)(tmpres % 1000000UL);
+    }
+
+    return 0;
+}
+
+#endif
+
+#if !defined(__WIN32__) && !defined(__WIN64__) && !defined(__CYGWIN32__) && 0
+
+static void *huge_malloc(BLASLONG size){
+  int shmid;
+  void *address;
+
+#ifndef SHM_HUGETLB
+#define SHM_HUGETLB 04000
+#endif
+
+  if ((shmid =shmget(IPC_PRIVATE,
+		     (size + HUGE_PAGESIZE) & ~(HUGE_PAGESIZE - 1),
+		     SHM_HUGETLB | IPC_CREAT |0600)) < 0) {
+    printf( "Memory allocation failed(shmget).\n");
+    exit(1);
+  }
+
+  address = shmat(shmid, NULL, SHM_RND);
+
+  if ((BLASLONG)address == -1){
+    printf( "Memory allocation failed(shmat).\n");
+    exit(1);
+  }
+
+  shmctl(shmid, IPC_RMID, 0);
+
+  return address;
+}
+
+#define malloc huge_malloc
+
+#endif
+
+int main(int argc, char *argv[]){
+
+    FLOAT *a, *x;
+    FLOAT alpha[] = {1.0, 1.0};
+    blasint incx = 1;
+    char *p;
+
+    char uplo='U';
+    char trans='N';
+
+    if ((p = getenv("OPENBLAS_UPLO"))) uplo=*p;
+    if ((p = getenv("OPENBLAS_TRANS"))) trans=*p;
+
+    blasint m, i, j;
+
+    int from =   1;
+    int to   = 200;
+    int step =   1;
+
+    struct timeval start, stop;
+    double time1;
+
+    argc--;argv++;
+
+    if (argc > 0) { from     = atol(*argv);		argc--; argv++;}
+    if (argc > 0) { to       = MAX(atol(*argv), from);	argc--; argv++;}
+    if (argc > 0) { step     = atol(*argv);		argc--; argv++;}
+
+    fprintf(stderr, "From : %3d  To : %3d Step = %3d Uplo = %c Trans = %c\n", from, to, step,uplo,trans);
+
+
+    if (( a = (FLOAT *)malloc(sizeof(FLOAT) * to * to * COMPSIZE)) == NULL){
+        fprintf(stderr,"Out of Memory!!\n");exit(1);
+    }
+
+    if (( x = (FLOAT *)malloc(sizeof(FLOAT) * to * COMPSIZE)) == NULL){
+        fprintf(stderr,"Out of Memory!!\n");exit(1);
+    }
+
+
+
+#ifdef linux
+    srandom(getpid());
+#endif
+
+    fprintf(stderr, "   SIZE       Flops\n");
+
+    for(m = from; m <= to; m += step)
+    {
+        fprintf(stderr, " %6d : ", (int)m);
+
+        for(j = 0; j < m; j++){
+            for(i = 0; i < m * COMPSIZE; i++){
+                a[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+            }
+            x[ (long)j * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+        }
+
+        gettimeofday( &start, (struct timezone *)0);
+
+        HER (&uplo, &m, alpha, x, &incx, a, &m );
+
+        gettimeofday( &stop, (struct timezone *)0);
+
+        time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
+
+        gettimeofday( &start, (struct timezone *)0);
+
+        fprintf(stderr,
+                " %10.2f MFlops\n",
+                COMPSIZE * COMPSIZE * 1. * (double)m * (double)m / time1 * 1.e-6);
+
+    }
+
+    return 0;
+}
diff --git a/benchmark/her2.c b/benchmark/her2.c
new file mode 100644
index 000000000..e10b7e98e
--- /dev/null
+++ b/benchmark/her2.c
@@ -0,0 +1,190 @@
+/***************************************************************************
+Copyright (c) 2020, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include <stdio.h>
+#include <stdlib.h>
+#ifdef __CYGWIN32__
+#include <sys/time.h>
+#endif
+#include "common.h"
+
+
+#undef HER2
+
+
+#ifdef DOUBLE
+#define HER2   BLASFUNC(zher2)
+#else
+#define HER2   BLASFUNC(cher2)
+#endif
+
+
+#if defined(__WIN32__) || defined(__WIN64__)
+
+#ifndef DELTA_EPOCH_IN_MICROSECS
+#define DELTA_EPOCH_IN_MICROSECS 11644473600000000ULL
+#endif
+
+int gettimeofday(struct timeval *tv, void *tz){
+
+    FILETIME ft;
+    unsigned __int64 tmpres = 0;
+    static int tzflag;
+
+    if (NULL != tv)
+    {
+        GetSystemTimeAsFileTime(&ft);
+
+        tmpres |= ft.dwHighDateTime;
+        tmpres <<= 32;
+        tmpres |= ft.dwLowDateTime;
+
+        /*converting file time to unix epoch*/
+        tmpres /= 10;  /*convert into microseconds*/
+        tmpres -= DELTA_EPOCH_IN_MICROSECS;
+        tv->tv_sec = (long)(tmpres / 1000000UL);
+        tv->tv_usec = (long)(tmpres % 1000000UL);
+    }
+
+    return 0;
+}
+
+#endif
+
+#if !defined(__WIN32__) && !defined(__WIN64__) && !defined(__CYGWIN32__) && 0
+
+static void *huge_malloc(BLASLONG size){
+  int shmid;
+  void *address;
+
+#ifndef SHM_HUGETLB
+#define SHM_HUGETLB 04000
+#endif
+
+  if ((shmid =shmget(IPC_PRIVATE,
+		     (size + HUGE_PAGESIZE) & ~(HUGE_PAGESIZE - 1),
+		     SHM_HUGETLB | IPC_CREAT |0600)) < 0) {
+    printf( "Memory allocation failed(shmget).\n");
+    exit(1);
+  }
+
+  address = shmat(shmid, NULL, SHM_RND);
+
+  if ((BLASLONG)address == -1){
+    printf( "Memory allocation failed(shmat).\n");
+    exit(1);
+  }
+
+  shmctl(shmid, IPC_RMID, 0);
+
+  return address;
+}
+
+#define malloc huge_malloc
+
+#endif
+
+int main(int argc, char *argv[]){
+
+    FLOAT *a, *x, *y;
+    FLOAT alpha[] = {1.0, 1.0};
+    blasint inc = 1;
+    char *p;
+
+    char uplo='U';
+    char trans='N';
+
+    if ((p = getenv("OPENBLAS_UPLO"))) uplo=*p;
+    if ((p = getenv("OPENBLAS_TRANS"))) trans=*p;
+
+    blasint m, i, j;
+
+    int from =   1;
+    int to   = 200;
+    int step =   1;
+
+    struct timeval start, stop;
+    double time1;
+
+    argc--;argv++;
+
+    if (argc > 0) { from     = atol(*argv);		argc--; argv++;}
+    if (argc > 0) { to       = MAX(atol(*argv), from);	argc--; argv++;}
+    if (argc > 0) { step     = atol(*argv);		argc--; argv++;}
+
+    fprintf(stderr, "From : %3d  To : %3d Step = %3d Uplo = %c Trans = %c\n", from, to, step,uplo,trans);
+
+
+    if (( a = (FLOAT *)malloc(sizeof(FLOAT) * to * to * COMPSIZE)) == NULL){
+        fprintf(stderr,"Out of Memory!!\n");exit(1);
+    }
+
+    if (( x = (FLOAT *)malloc(sizeof(FLOAT) * to * COMPSIZE)) == NULL){
+        fprintf(stderr,"Out of Memory!!\n");exit(1);
+    }
+
+    if (( y = (FLOAT *)malloc(sizeof(FLOAT) * to * COMPSIZE)) == NULL){
+        fprintf(stderr,"Out of Memory!!\n");exit(1);
+    }
+
+#ifdef linux
+    srandom(getpid());
+#endif
+
+    fprintf(stderr, "   SIZE       Flops\n");
+
+    for(m = from; m <= to; m += step)
+    {
+        fprintf(stderr, " %6d : ", (int)m);
+
+        for(j = 0; j < m; j++){
+            for(i = 0; i < m * COMPSIZE; i++){
+                a[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+            }
+            x[ (long)j * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+            y[ (long)j * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+        }
+
+        gettimeofday( &start, (struct timezone *)0);
+
+
+        HER2 (&uplo, &m, alpha, x, &inc, y, &inc, a, &m );
+
+        gettimeofday( &stop, (struct timezone *)0);
+
+        time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
+
+        gettimeofday( &start, (struct timezone *)0);
+
+        fprintf(stderr,
+                " %10.2f MFlops\n",
+                COMPSIZE * COMPSIZE * 2. * (double)m * (double)m / time1 * 1.e-6);
+
+    }
+
+    return 0;
+}
diff --git a/benchmark/her2k.c b/benchmark/her2k.c
index 028e2718f..a0772feff 100644
--- a/benchmark/her2k.c
+++ b/benchmark/her2k.c
@@ -163,9 +163,9 @@ int main(int argc, char *argv[]){
 
     for(j = 0; j < m; j++){
       for(i = 0; i < m * COMPSIZE; i++){
-	a[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
-	b[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
-	c[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+	a[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+	b[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+	c[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
       }
     }
 
@@ -177,8 +177,6 @@ int main(int argc, char *argv[]){
 
     time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
 
-    gettimeofday( &start, (struct timezone *)0);
-
     fprintf(stderr,
 	    " %10.2f MFlops\n",
 	    COMPSIZE * COMPSIZE * 2. * (double)m * (double)m * (double)m / time1 * 1.e-6);
diff --git a/benchmark/herk.c b/benchmark/herk.c
index d2e25ff46..eed8ed738 100644
--- a/benchmark/herk.c
+++ b/benchmark/herk.c
@@ -162,8 +162,8 @@ int main(int argc, char *argv[]){
 
     for(j = 0; j < m; j++){
       for(i = 0; i < m * COMPSIZE; i++){
-	a[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
-	c[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+	a[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+	c[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
       }
     }
 
@@ -175,8 +175,6 @@ int main(int argc, char *argv[]){
 
     time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
 
-    gettimeofday( &start, (struct timezone *)0);
-
     fprintf(stderr,
 	    " %10.2f MFlops\n",
 	    COMPSIZE * COMPSIZE * 1. * (double)m * (double)m * (double)m / time1 * 1.e-6);
diff --git a/benchmark/hpmv.c b/benchmark/hpmv.c
new file mode 100644
index 000000000..6e6634fcf
--- /dev/null
+++ b/benchmark/hpmv.c
@@ -0,0 +1,207 @@
+/***************************************************************************
+Copyright (c) 2014, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include <stdio.h>
+#include <stdlib.h>
+#ifdef __CYGWIN32__
+#include <sys/time.h>
+#endif
+#include "common.h"
+
+
+#undef HPMV
+
+
+#ifdef DOUBLE
+#define HPMV   BLASFUNC(zhpmv)
+#else
+#define HPMV   BLASFUNC(chpmv)
+#endif
+
+
+#if defined(__WIN32__) || defined(__WIN64__)
+
+#ifndef DELTA_EPOCH_IN_MICROSECS
+#define DELTA_EPOCH_IN_MICROSECS 11644473600000000ULL
+#endif
+
+int gettimeofday(struct timeval *tv, void *tz) {
+
+    FILETIME ft;
+    unsigned __int64 tmpres = 0;
+    static int tzflag;
+
+    if (NULL != tv)
+    {
+        GetSystemTimeAsFileTime(&ft);
+
+        tmpres |= ft.dwHighDateTime;
+        tmpres <<= 32;
+        tmpres |= ft.dwLowDateTime;
+
+        /*converting file time to unix epoch*/
+        tmpres /= 10;  /*convert into microseconds*/
+        tmpres -= DELTA_EPOCH_IN_MICROSECS;
+        tv->tv_sec = (long)(tmpres / 1000000UL);
+        tv->tv_usec = (long)(tmpres % 1000000UL);
+    }
+
+    return 0;
+}
+
+#endif
+
+#if !defined(__WIN32__) && !defined(__WIN64__) && !defined(__CYGWIN32__) && 0
+
+static void *huge_malloc(BLASLONG size) {
+    int shmid;
+    void *address;
+
+#ifndef SHM_HUGETLB
+#define SHM_HUGETLB 04000
+#endif
+
+    if ((shmid =shmget(IPC_PRIVATE,
+                (size + HUGE_PAGESIZE) & ~(HUGE_PAGESIZE - 1),
+                SHM_HUGETLB | IPC_CREAT |0600)) < 0) {
+        printf( "Memory allocation failed(shmget).\n");
+        exit(1);
+    }
+
+    address = shmat(shmid, NULL, SHM_RND);
+
+    if ((BLASLONG)address == -1){
+        printf( "Memory allocation failed(shmat).\n");
+        exit(1);
+    }
+
+    shmctl(shmid, IPC_RMID, 0);
+
+    return address;
+}
+
+#define malloc huge_malloc
+
+#endif
+
+int main(int argc, char *argv[]){
+
+    FLOAT *a, *x, *y;
+    FLOAT alpha[] = {1.0, 1.0};
+    FLOAT beta [] = {1.0, 1.0};
+    char uplo='L';
+    blasint m, i, j;
+    blasint inc_x=1, inc_y=1;
+    int loops = 1;
+    int l;
+    char *p;
+
+    int from =   1;
+    int to   = 200;
+    int step =   1;
+
+    struct timeval start, stop;
+    double time1,timeg;
+
+    argc--;argv++;
+
+    if (argc > 0) { from     = atol(*argv);		argc--; argv++;}
+    if (argc > 0) { to       = MAX(atol(*argv), from);	argc--; argv++;}
+    if (argc > 0) { step     = atol(*argv);		argc--; argv++;}
+
+    if ((p = getenv("OPENBLAS_LOOPS")))  loops = atoi(p);
+    if ((p = getenv("OPENBLAS_INCX")))   inc_x = atoi(p);
+    if ((p = getenv("OPENBLAS_INCY")))   inc_y = atoi(p);
+    if ((p = getenv("OPENBLAS_UPLO")))   uplo=*p;
+
+    fprintf(stderr, "From : %3d  To : %3d Step = %3d Uplo = '%c' Inc_x = %d Inc_y = %d Loops = %d\n", from, to, step,uplo,inc_x,inc_y,loops);
+
+    if (( a = (FLOAT *)malloc(sizeof(FLOAT) * to * to * COMPSIZE)) == NULL) {
+        fprintf(stderr,"Out of Memory!!\n");
+        exit(1);
+    }
+
+    if (( x = (FLOAT *)malloc(sizeof(FLOAT) * to * abs(inc_x) * COMPSIZE)) == NULL) {
+        fprintf(stderr,"Out of Memory!!\n");
+        exit(1);
+    }
+
+    if (( y = (FLOAT *)malloc(sizeof(FLOAT) * to * abs(inc_y) * COMPSIZE)) == NULL) {
+        fprintf(stderr,"Out of Memory!!\n");
+        exit(1);
+    }
+
+#ifdef linux
+    srandom(getpid());
+#endif
+
+    fprintf(stderr, "   SIZE       Flops\n");
+
+    for(m = from; m <= to; m += step) {
+
+        timeg=0;
+
+        fprintf(stderr, " %6dx%d : ", (int)m, (int)m);
+
+        for(j = 0; j < m; j++) {
+            for(i = 0; i < m * COMPSIZE; i++) {
+                a[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+            }
+        }
+
+    for (l = 0; l < loops; l++) {
+
+        for (i = 0; i < m * COMPSIZE * abs(inc_x); i++) {
+            x[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+        }
+
+        for (i = 0; i < m * COMPSIZE * abs(inc_y); i++) {
+            y[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+        }
+
+        gettimeofday( &start, (struct timezone *)0);
+
+        HPMV (&uplo, &m, alpha, a, x, &inc_x, beta, y, &inc_y );
+
+        gettimeofday( &stop, (struct timezone *)0);
+
+        time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
+
+        timeg += time1;
+
+    }
+
+    timeg /= loops;
+
+    fprintf(stderr, " %10.2f MFlops\n",
+            COMPSIZE * COMPSIZE * 2. * (double)m * (double)m / timeg * 1.e-6);
+    }
+
+    return 0;
+}
+
+// void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));
diff --git a/benchmark/iamax.c b/benchmark/iamax.c
index 034e24ea9..736f02b89 100644
--- a/benchmark/iamax.c
+++ b/benchmark/iamax.c
@@ -181,7 +181,7 @@ int main(int argc, char *argv[]){
     timeg /= loops;
 
     fprintf(stderr,
-	    " %10.2f MFlops %10.6f sec\n",
+	    " %10.2f MBytes %10.6f sec\n",
 	    COMPSIZE * sizeof(FLOAT) * 1. * (double)m / timeg * 1.e-6, timeg);
 
   }
diff --git a/benchmark/iamin.c b/benchmark/iamin.c
new file mode 100644
index 000000000..b2c779811
--- /dev/null
+++ b/benchmark/iamin.c
@@ -0,0 +1,192 @@
+/***************************************************************************
+Copyright (c) 2016, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include <stdio.h>
+#include <stdlib.h>
+#ifdef __CYGWIN32__
+#include <sys/time.h>
+#endif
+#include "common.h"
+
+
+#undef IAMIN
+
+#ifdef COMPLEX
+#ifdef DOUBLE
+#define IAMIN   BLASFUNC(izamin)
+#else
+#define IAMIN   BLASFUNC(icamin)
+#endif
+#else
+#ifdef DOUBLE
+#define IAMIN   BLASFUNC(idamin)
+#else
+#define IAMIN   BLASFUNC(isamin)
+#endif
+#endif
+
+#if defined(__WIN32__) || defined(__WIN64__)
+
+#ifndef DELTA_EPOCH_IN_MICROSECS
+#define DELTA_EPOCH_IN_MICROSECS 11644473600000000ULL
+#endif
+
+int gettimeofday(struct timeval *tv, void *tz){
+
+  FILETIME ft;
+  unsigned __int64 tmpres = 0;
+  static int tzflag;
+
+  if (NULL != tv)
+    {
+      GetSystemTimeAsFileTime(&ft);
+
+      tmpres |= ft.dwHighDateTime;
+      tmpres <<= 32;
+      tmpres |= ft.dwLowDateTime;
+
+      /*converting file time to unix epoch*/
+      tmpres /= 10;  /*convert into microseconds*/
+      tmpres -= DELTA_EPOCH_IN_MICROSECS;
+      tv->tv_sec = (long)(tmpres / 1000000UL);
+      tv->tv_usec = (long)(tmpres % 1000000UL);
+    }
+
+  return 0;
+}
+
+#endif
+
+#if !defined(__WIN32__) && !defined(__WIN64__) && !defined(__CYGWIN32__) && 0
+
+static void *huge_malloc(BLASLONG size){
+  int shmid;
+  void *address;
+
+#ifndef SHM_HUGETLB
+#define SHM_HUGETLB 04000
+#endif
+
+  if ((shmid =shmget(IPC_PRIVATE,
+		     (size + HUGE_PAGESIZE) & ~(HUGE_PAGESIZE - 1),
+		     SHM_HUGETLB | IPC_CREAT |0600)) < 0) {
+    printf( "Memory allocation failed(shmget).\n");
+    exit(1);
+  }
+
+  address = shmat(shmid, NULL, SHM_RND);
+
+  if ((BLASLONG)address == -1){
+    printf( "Memory allocation failed(shmat).\n");
+    exit(1);
+  }
+
+  shmctl(shmid, IPC_RMID, 0);
+
+  return address;
+}
+
+#define malloc huge_malloc
+
+#endif
+
+int main(int argc, char *argv[]){
+
+  FLOAT *x;
+  blasint m, i;
+  blasint inc_x=1;
+  int loops = 1;
+  int l;
+  char *p;
+
+  int from =   1;
+  int to   = 200;
+  int step =   1;
+
+  struct timeval start, stop;
+  double time1,timeg;
+
+  argc--;argv++;
+
+  if (argc > 0) { from     = atol(*argv);		argc--; argv++;}
+  if (argc > 0) { to       = MAX(atol(*argv), from);	argc--; argv++;}
+  if (argc > 0) { step     = atol(*argv);		argc--; argv++;}
+
+  if ((p = getenv("OPENBLAS_LOOPS")))  loops = atoi(p);
+  if ((p = getenv("OPENBLAS_INCX")))   inc_x = atoi(p);
+
+  fprintf(stderr, "From : %3d  To : %3d Step = %3d Inc_x = %d Loops = %d\n", from, to, step,inc_x,loops);
+
+  if (( x = (FLOAT *)malloc(sizeof(FLOAT) * to * abs(inc_x) * COMPSIZE)) == NULL){
+    fprintf(stderr,"Out of Memory!!\n");exit(1);
+  }
+
+#ifdef linux
+  srandom(getpid());
+#endif
+
+  fprintf(stderr, "   SIZE       Flops\n");
+
+  for(m = from; m <= to; m += step)
+  {
+
+   timeg=0;
+
+   fprintf(stderr, " %6d : ", (int)m);
+
+
+   for (l=0; l<loops; l++)
+   {
+
+   	for(i = 0; i < m * COMPSIZE * abs(inc_x); i++){
+			x[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+   	}
+
+    	gettimeofday( &start, (struct timezone *)0);
+
+    	IAMIN (&m, x, &inc_x);
+
+    	gettimeofday( &stop, (struct timezone *)0);
+
+    	time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
+
+	timeg += time1;
+
+    }
+
+    timeg /= loops;
+
+    fprintf(stderr,
+	    " %10.2f MFlops %10.6f sec\n",
+	    COMPSIZE * sizeof(FLOAT) * 1. * (double)m / timeg * 1.e-6, timeg);
+
+  }
+  
+  return 0;
+}
+
+// void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));
diff --git a/benchmark/imax.c b/benchmark/imax.c
new file mode 100644
index 000000000..c7060af84
--- /dev/null
+++ b/benchmark/imax.c
@@ -0,0 +1,186 @@
+/***************************************************************************
+Copyright (c) 2016, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include <stdio.h>
+#include <stdlib.h>
+#ifdef __CYGWIN32__
+#include <sys/time.h>
+#endif
+#include "common.h"
+
+
+#undef IMAX
+
+#ifndef COMPLEX
+#ifdef DOUBLE
+#define IMAX   BLASFUNC(idmax)
+#else
+#define IMAX   BLASFUNC(ismax)
+#endif
+#endif
+
+#if defined(__WIN32__) || defined(__WIN64__)
+
+#ifndef DELTA_EPOCH_IN_MICROSECS
+#define DELTA_EPOCH_IN_MICROSECS 11644473600000000ULL
+#endif
+
+int gettimeofday(struct timeval *tv, void *tz){
+
+  FILETIME ft;
+  unsigned __int64 tmpres = 0;
+  static int tzflag;
+
+  if (NULL != tv)
+    {
+      GetSystemTimeAsFileTime(&ft);
+
+      tmpres |= ft.dwHighDateTime;
+      tmpres <<= 32;
+      tmpres |= ft.dwLowDateTime;
+
+      /*converting file time to unix epoch*/
+      tmpres /= 10;  /*convert into microseconds*/
+      tmpres -= DELTA_EPOCH_IN_MICROSECS;
+      tv->tv_sec = (long)(tmpres / 1000000UL);
+      tv->tv_usec = (long)(tmpres % 1000000UL);
+    }
+
+  return 0;
+}
+
+#endif
+
+#if !defined(__WIN32__) && !defined(__WIN64__) && !defined(__CYGWIN32__) && 0
+
+static void *huge_malloc(BLASLONG size){
+  int shmid;
+  void *address;
+
+#ifndef SHM_HUGETLB
+#define SHM_HUGETLB 04000
+#endif
+
+  if ((shmid =shmget(IPC_PRIVATE,
+		     (size + HUGE_PAGESIZE) & ~(HUGE_PAGESIZE - 1),
+		     SHM_HUGETLB | IPC_CREAT |0600)) < 0) {
+    printf( "Memory allocation failed(shmget).\n");
+    exit(1);
+  }
+
+  address = shmat(shmid, NULL, SHM_RND);
+
+  if ((BLASLONG)address == -1){
+    printf( "Memory allocation failed(shmat).\n");
+    exit(1);
+  }
+
+  shmctl(shmid, IPC_RMID, 0);
+
+  return address;
+}
+
+#define malloc huge_malloc
+
+#endif
+
+int main(int argc, char *argv[]){
+
+  FLOAT *x;
+  blasint m, i;
+  blasint inc_x=1;
+  int loops = 1;
+  int l;
+  char *p;
+
+  int from =   1;
+  int to   = 200;
+  int step =   1;
+
+  struct timeval start, stop;
+  double time1,timeg;
+
+  argc--;argv++;
+
+  if (argc > 0) { from     = atol(*argv);		argc--; argv++;}
+  if (argc > 0) { to       = MAX(atol(*argv), from);	argc--; argv++;}
+  if (argc > 0) { step     = atol(*argv);		argc--; argv++;}
+
+  if ((p = getenv("OPENBLAS_LOOPS")))  loops = atoi(p);
+  if ((p = getenv("OPENBLAS_INCX")))   inc_x = atoi(p);
+
+  fprintf(stderr, "From : %3d  To : %3d Step = %3d Inc_x = %d Loops = %d\n", from, to, step,inc_x,loops);
+
+  if (( x = (FLOAT *)malloc(sizeof(FLOAT) * to * abs(inc_x) * COMPSIZE)) == NULL){
+    fprintf(stderr,"Out of Memory!!\n");exit(1);
+  }
+
+#ifdef linux
+  srandom(getpid());
+#endif
+
+  fprintf(stderr, "   SIZE       Flops\n");
+
+  for(m = from; m <= to; m += step)
+  {
+
+   timeg=0;
+
+   fprintf(stderr, " %6d : ", (int)m);
+
+
+   for (l=0; l<loops; l++)
+   {
+
+   	for(i = 0; i < m * COMPSIZE * abs(inc_x); i++){
+			x[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+   	}
+
+    	gettimeofday( &start, (struct timezone *)0);
+
+    	IMAX (&m, x, &inc_x);
+
+    	gettimeofday( &stop, (struct timezone *)0);
+
+    	time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
+
+	timeg += time1;
+
+    }
+
+    timeg /= loops;
+
+    fprintf(stderr,
+	    " %10.2f MFlops %10.6f sec\n",
+	    COMPSIZE * sizeof(FLOAT) * 1. * (double)m / timeg * 1.e-6, timeg);
+
+  }
+
+  return 0;
+}
+
+// void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));
diff --git a/benchmark/imin.c b/benchmark/imin.c
new file mode 100644
index 000000000..f8bdc2537
--- /dev/null
+++ b/benchmark/imin.c
@@ -0,0 +1,186 @@
+/***************************************************************************
+Copyright (c) 2016, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include <stdio.h>
+#include <stdlib.h>
+#ifdef __CYGWIN32__
+#include <sys/time.h>
+#endif
+#include "common.h"
+
+
+#undef IMIN
+
+#ifndef COMPLEX
+#ifdef DOUBLE
+#define IMIN   BLASFUNC(idmin)
+#else
+#define IMIN   BLASFUNC(ismin)
+#endif
+#endif
+
+#if defined(__WIN32__) || defined(__WIN64__)
+
+#ifndef DELTA_EPOCH_IN_MICROSECS
+#define DELTA_EPOCH_IN_MICROSECS 11644473600000000ULL
+#endif
+
+int gettimeofday(struct timeval *tv, void *tz){
+
+  FILETIME ft;
+  unsigned __int64 tmpres = 0;
+  static int tzflag;
+
+  if (NULL != tv)
+    {
+      GetSystemTimeAsFileTime(&ft);
+
+      tmpres |= ft.dwHighDateTime;
+      tmpres <<= 32;
+      tmpres |= ft.dwLowDateTime;
+
+      /*converting file time to unix epoch*/
+      tmpres /= 10;  /*convert into microseconds*/
+      tmpres -= DELTA_EPOCH_IN_MICROSECS;
+      tv->tv_sec = (long)(tmpres / 1000000UL);
+      tv->tv_usec = (long)(tmpres % 1000000UL);
+    }
+
+  return 0;
+}
+
+#endif
+
+#if !defined(__WIN32__) && !defined(__WIN64__) && !defined(__CYGWIN32__) && 0
+
+static void *huge_malloc(BLASLONG size){
+  int shmid;
+  void *address;
+
+#ifndef SHM_HUGETLB
+#define SHM_HUGETLB 04000
+#endif
+
+  if ((shmid =shmget(IPC_PRIVATE,
+		     (size + HUGE_PAGESIZE) & ~(HUGE_PAGESIZE - 1),
+		     SHM_HUGETLB | IPC_CREAT |0600)) < 0) {
+    printf( "Memory allocation failed(shmget).\n");
+    exit(1);
+  }
+
+  address = shmat(shmid, NULL, SHM_RND);
+
+  if ((BLASLONG)address == -1){
+    printf( "Memory allocation failed(shmat).\n");
+    exit(1);
+  }
+
+  shmctl(shmid, IPC_RMID, 0);
+
+  return address;
+}
+
+#define malloc huge_malloc
+
+#endif
+
+int main(int argc, char *argv[]){
+
+  FLOAT *x;
+  blasint m, i;
+  blasint inc_x=1;
+  int loops = 1;
+  int l;
+  char *p;
+
+  int from =   1;
+  int to   = 200;
+  int step =   1;
+
+  struct timeval start, stop;
+  double time1,timeg;
+
+  argc--;argv++;
+
+  if (argc > 0) { from     = atol(*argv);		argc--; argv++;}
+  if (argc > 0) { to       = MAX(atol(*argv), from);	argc--; argv++;}
+  if (argc > 0) { step     = atol(*argv);		argc--; argv++;}
+
+  if ((p = getenv("OPENBLAS_LOOPS")))  loops = atoi(p);
+  if ((p = getenv("OPENBLAS_INCX")))   inc_x = atoi(p);
+
+  fprintf(stderr, "From : %3d  To : %3d Step = %3d Inc_x = %d Loops = %d\n", from, to, step,inc_x,loops);
+
+  if (( x = (FLOAT *)malloc(sizeof(FLOAT) * to * abs(inc_x) * COMPSIZE)) == NULL){
+    fprintf(stderr,"Out of Memory!!\n");exit(1);
+  }
+
+#ifdef linux
+  srandom(getpid());
+#endif
+
+  fprintf(stderr, "   SIZE       Flops\n");
+
+  for(m = from; m <= to; m += step)
+  {
+
+   timeg=0;
+
+   fprintf(stderr, " %6d : ", (int)m);
+
+
+   for (l=0; l<loops; l++)
+   {
+
+   	for(i = 0; i < m * COMPSIZE * abs(inc_x); i++){
+			x[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+   	}
+
+    	gettimeofday( &start, (struct timezone *)0);
+
+    	IMIN (&m, x, &inc_x);
+
+    	gettimeofday( &stop, (struct timezone *)0);
+
+    	time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
+
+	timeg += time1;
+
+    }
+
+    timeg /= loops;
+
+    fprintf(stderr,
+	    " %10.2f MFlops %10.6f sec\n",
+	    COMPSIZE * sizeof(FLOAT) * 1. * (double)m / timeg * 1.e-6, timeg);
+
+  }
+
+  return 0;
+}
+
+// void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));
diff --git a/benchmark/linpack.c b/benchmark/linpack.c
index 7d5c87163..e4b20e99d 100644
--- a/benchmark/linpack.c
+++ b/benchmark/linpack.c
@@ -186,7 +186,7 @@ int main(int argc, char *argv[]){
 
     for(j = 0; j < m; j++){
       for(i = 0; i < m * COMPSIZE; i++){
-	a[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+	a[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
       }
     }
 
@@ -194,7 +194,7 @@ int main(int argc, char *argv[]){
 
     for (j = 0; j < m; ++j) {
       for (i = 0; i < m * COMPSIZE; ++i) {
-	b[i] += a[i + j * m * COMPSIZE];
+	b[i] += a[(long)i + (long)j * (long)m * COMPSIZE];
       }
     }
 
diff --git a/benchmark/max.c b/benchmark/max.c
new file mode 100644
index 000000000..2fa6e5a14
--- /dev/null
+++ b/benchmark/max.c
@@ -0,0 +1,185 @@
+/***************************************************************************
+Copyright (c) 2016, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include <stdio.h>
+#include <stdlib.h>
+#ifdef __CYGWIN32__
+#include <sys/time.h>
+#endif
+#include "common.h"
+
+
+#undef NAMAX
+
+#ifndef COMPLEX
+#ifdef DOUBLE
+#define NAMAX   BLASFUNC(dmax)
+#else
+#define NAMAX   BLASFUNC(smax)
+#endif
+#endif
+
+#if defined(__WIN32__) || defined(__WIN64__)
+
+#ifndef DELTA_EPOCH_IN_MICROSECS
+#define DELTA_EPOCH_IN_MICROSECS 11644473600000000ULL
+#endif
+
+int gettimeofday(struct timeval *tv, void *tz){
+
+  FILETIME ft;
+  unsigned __int64 tmpres = 0;
+  static int tzflag;
+
+  if (NULL != tv)
+    {
+      GetSystemTimeAsFileTime(&ft);
+
+      tmpres |= ft.dwHighDateTime;
+      tmpres <<= 32;
+      tmpres |= ft.dwLowDateTime;
+
+      /*converting file time to unix epoch*/
+      tmpres /= 10;  /*convert into microseconds*/
+      tmpres -= DELTA_EPOCH_IN_MICROSECS;
+      tv->tv_sec = (long)(tmpres / 1000000UL);
+      tv->tv_usec = (long)(tmpres % 1000000UL);
+    }
+
+  return 0;
+}
+
+#endif
+
+#if !defined(__WIN32__) && !defined(__WIN64__) && !defined(__CYGWIN32__) && 0
+
+static void *huge_malloc(BLASLONG size){
+  int shmid;
+  void *address;
+
+#ifndef SHM_HUGETLB
+#define SHM_HUGETLB 04000
+#endif
+
+  if ((shmid =shmget(IPC_PRIVATE,
+		     (size + HUGE_PAGESIZE) & ~(HUGE_PAGESIZE - 1),
+		     SHM_HUGETLB | IPC_CREAT |0600)) < 0) {
+    printf( "Memory allocation failed(shmget).\n");
+    exit(1);
+  }
+
+  address = shmat(shmid, NULL, SHM_RND);
+
+  if ((BLASLONG)address == -1){
+    printf( "Memory allocation failed(shmat).\n");
+    exit(1);
+  }
+
+  shmctl(shmid, IPC_RMID, 0);
+
+  return address;
+}
+
+#define malloc huge_malloc
+
+#endif
+
+int main(int argc, char *argv[]){
+
+  FLOAT *x;
+  blasint m, i;
+  blasint inc_x=1;
+  int loops = 1;
+  int l;
+  char *p;
+
+  int from =   1;
+  int to   = 200;
+  int step =   1;
+
+  struct timeval start, stop;
+  double time1,timeg;
+
+  argc--;argv++;
+
+  if (argc > 0) { from     = atol(*argv);		argc--; argv++;}
+  if (argc > 0) { to       = MAX(atol(*argv), from);	argc--; argv++;}
+  if (argc > 0) { step     = atol(*argv);		argc--; argv++;}
+
+  if ((p = getenv("OPENBLAS_LOOPS")))  loops = atoi(p);
+  if ((p = getenv("OPENBLAS_INCX")))   inc_x = atoi(p);
+
+  fprintf(stderr, "From : %3d  To : %3d Step = %3d Inc_x = %d Loops = %d\n", from, to, step,inc_x,loops);
+
+  if (( x = (FLOAT *)malloc(sizeof(FLOAT) * to * abs(inc_x) * COMPSIZE)) == NULL){
+    fprintf(stderr,"Out of Memory!!\n");exit(1);
+  }
+
+#ifdef linux
+  srandom(getpid());
+#endif
+
+  fprintf(stderr, "   SIZE       Flops\n");
+
+  for(m = from; m <= to; m += step)
+  {
+
+   timeg=0;
+
+   fprintf(stderr, " %6d : ", (int)m);
+
+
+   for (l=0; l<loops; l++)
+   {
+
+   	for(i = 0; i < m * COMPSIZE * abs(inc_x); i++){
+			x[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+   	}
+
+    	gettimeofday( &start, (struct timezone *)0);
+
+    	NAMAX (&m, x, &inc_x);
+
+    	gettimeofday( &stop, (struct timezone *)0);
+
+    	time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
+
+	timeg += time1;
+
+    }
+
+    timeg /= loops;
+
+    fprintf(stderr,
+	    " %10.2f MFlops %10.6f sec\n",
+	    COMPSIZE * sizeof(FLOAT) * 1. * (double)m / timeg * 1.e-6, timeg);
+  }
+
+  return 0;
+}
+
+// void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));
diff --git a/benchmark/min.c b/benchmark/min.c
new file mode 100644
index 000000000..9abed8e80
--- /dev/null
+++ b/benchmark/min.c
@@ -0,0 +1,185 @@
+/***************************************************************************
+Copyright (c) 2016, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include <stdio.h>
+#include <stdlib.h>
+#ifdef __CYGWIN32__
+#include <sys/time.h>
+#endif
+#include "common.h"
+
+
+#undef NAMIN
+
+#ifndef COMPLEX
+#ifdef DOUBLE
+#define NAMIN   BLASFUNC(dmin)
+#else
+#define NAMIN   BLASFUNC(smin)
+#endif
+#endif
+
+#if defined(__WIN32__) || defined(__WIN64__)
+
+#ifndef DELTA_EPOCH_IN_MICROSECS
+#define DELTA_EPOCH_IN_MICROSECS 11644473600000000ULL
+#endif
+
+int gettimeofday(struct timeval *tv, void *tz){
+
+  FILETIME ft;
+  unsigned __int64 tmpres = 0;
+  static int tzflag;
+
+  if (NULL != tv)
+    {
+      GetSystemTimeAsFileTime(&ft);
+
+      tmpres |= ft.dwHighDateTime;
+      tmpres <<= 32;
+      tmpres |= ft.dwLowDateTime;
+
+      /*converting file time to unix epoch*/
+      tmpres /= 10;  /*convert into microseconds*/
+      tmpres -= DELTA_EPOCH_IN_MICROSECS;
+      tv->tv_sec = (long)(tmpres / 1000000UL);
+      tv->tv_usec = (long)(tmpres % 1000000UL);
+    }
+
+  return 0;
+}
+
+#endif
+
+#if !defined(__WIN32__) && !defined(__WIN64__) && !defined(__CYGWIN32__) && 0
+
+static void *huge_malloc(BLASLONG size){
+  int shmid;
+  void *address;
+
+#ifndef SHM_HUGETLB
+#define SHM_HUGETLB 04000
+#endif
+
+  if ((shmid =shmget(IPC_PRIVATE,
+		     (size + HUGE_PAGESIZE) & ~(HUGE_PAGESIZE - 1),
+		     SHM_HUGETLB | IPC_CREAT |0600)) < 0) {
+    printf( "Memory allocation failed(shmget).\n");
+    exit(1);
+  }
+
+  address = shmat(shmid, NULL, SHM_RND);
+
+  if ((BLASLONG)address == -1){
+    printf( "Memory allocation failed(shmat).\n");
+    exit(1);
+  }
+
+  shmctl(shmid, IPC_RMID, 0);
+
+  return address;
+}
+
+#define malloc huge_malloc
+
+#endif
+
+int main(int argc, char *argv[]){
+
+  FLOAT *x;
+  blasint m, i;
+  blasint inc_x=1;
+  int loops = 1;
+  int l;
+  char *p;
+
+  int from =   1;
+  int to   = 200;
+  int step =   1;
+
+  struct timeval start, stop;
+  double time1,timeg;
+
+  argc--;argv++;
+
+  if (argc > 0) { from     = atol(*argv);		argc--; argv++;}
+  if (argc > 0) { to       = MAX(atol(*argv), from);	argc--; argv++;}
+  if (argc > 0) { step     = atol(*argv);		argc--; argv++;}
+
+  if ((p = getenv("OPENBLAS_LOOPS")))  loops = atoi(p);
+  if ((p = getenv("OPENBLAS_INCX")))   inc_x = atoi(p);
+
+  fprintf(stderr, "From : %3d  To : %3d Step = %3d Inc_x = %d Loops = %d\n", from, to, step,inc_x,loops);
+
+  if (( x = (FLOAT *)malloc(sizeof(FLOAT) * to * abs(inc_x) * COMPSIZE)) == NULL){
+    fprintf(stderr,"Out of Memory!!\n");exit(1);
+  }
+
+#ifdef linux
+  srandom(getpid());
+#endif
+
+  fprintf(stderr, "   SIZE       Flops\n");
+
+  for(m = from; m <= to; m += step)
+  {
+
+   timeg=0;
+
+   fprintf(stderr, " %6d : ", (int)m);
+
+
+   for (l=0; l<loops; l++)
+   {
+
+   	for(i = 0; i < m * COMPSIZE * abs(inc_x); i++){
+			x[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+   	}
+
+    	gettimeofday( &start, (struct timezone *)0);
+
+    	NAMIN (&m, x, &inc_x);
+
+    	gettimeofday( &stop, (struct timezone *)0);
+
+    	time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
+
+	timeg += time1;
+
+    }
+
+    timeg /= loops;
+
+    fprintf(stderr,
+	    " %10.2f MFlops %10.6f sec\n",
+	    COMPSIZE * sizeof(FLOAT) * 1. * (double)m / timeg * 1.e-6, timeg);
+  }
+
+  return 0;
+}
+
+// void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));
diff --git a/benchmark/potrf.c b/benchmark/potrf.c
index 1d714549b..cb4c23bab 100644
--- a/benchmark/potrf.c
+++ b/benchmark/potrf.c
@@ -170,46 +170,46 @@ int main(int argc, char *argv[]){
 #ifndef COMPLEX
       if (uplos & 1) {
 	for (j = 0; j < m; j++) {
-	  for(i = 0; i < j; i++)     a[i + j * m] = 0.;
-	                             a[j + j * m] = ((double) rand() / (double) RAND_MAX) + 8.;
-	  for(i = j + 1; i < m; i++) a[i + j * m] = ((double) rand() / (double) RAND_MAX) - 0.5;
+	  for(i = 0; i < j; i++)     a[(long)i + (long)j * (long)m] = 0.;
+	                             a[(long)j + (long)j * (long)m] = ((double) rand() / (double) RAND_MAX) + 8.;
+	  for(i = j + 1; i < m; i++) a[(long)i + (long)j * (long)m] = ((double) rand() / (double) RAND_MAX) - 0.5;
 	}
       } else {
 	for (j = 0; j < m; j++) {
-	  for(i = 0; i < j; i++)     a[i + j * m] = ((double) rand() / (double) RAND_MAX) - 0.5;
-	                             a[j + j * m] = ((double) rand() / (double) RAND_MAX) + 8.;
-	  for(i = j + 1; i < m; i++) a[i + j * m] = 0.;
+	  for(i = 0; i < j; i++)     a[(long)i + (long)j * (long)m] = ((double) rand() / (double) RAND_MAX) - 0.5;
+	                             a[(long)j + (long)j * (long)m] = ((double) rand() / (double) RAND_MAX) + 8.;
+	  for(i = j + 1; i < m; i++) a[(long)i + (long)j * (long)m] = 0.;
 	}
       }
 #else
       if (uplos & 1) {
 	for (j = 0; j < m; j++) {
 	  for(i = 0; i < j; i++) {
-	    a[(i + j * m) * 2 + 0] = 0.;
-	    a[(i + j * m) * 2 + 1] = 0.;
+	    a[((long)i + (long)j * (long)m) * 2 + 0] = 0.;
+	    a[((long)i + (long)j * (long)m) * 2 + 1] = 0.;
 	  }
 
-	  a[(j + j * m) * 2 + 0] = ((double) rand() / (double) RAND_MAX) + 8.;
-	  a[(j + j * m) * 2 + 1] = 0.;
+	  a[((long)j + (long)j * (long)m) * 2 + 0] = ((double) rand() / (double) RAND_MAX) + 8.;
+	  a[((long)j + (long)j * (long)m) * 2 + 1] = 0.;
 
 	  for(i = j + 1; i < m; i++) {
-	    a[(i + j * m) * 2 + 0] = ((double) rand() / (double) RAND_MAX) - 0.5;
-	    a[(i + j * m) * 2 + 1] = ((double) rand() / (double) RAND_MAX) - 0.5;
+	    a[((long)i + (long)j * (long)m) * 2 + 0] = 0;
+	    a[((long)i + (long)j * (long)m) * 2 + 1] = ((double) rand() / (double) RAND_MAX) - 0.5;
 	  }
 	}
       } else {
 	for (j = 0; j < m; j++) {
 	  for(i = 0; i < j; i++) {
-	    a[(i + j * m) * 2 + 0] = ((double) rand() / (double) RAND_MAX) - 0.5;
-	    a[(i + j * m) * 2 + 1] = ((double) rand() / (double) RAND_MAX) - 0.5;
+	    a[((long)i + (long)j * (long)m) * 2 + 0] = 0.;
+	    a[((long)i + (long)j * (long)m) * 2 + 1] = ((double) rand() / (double) RAND_MAX) - 0.5;
 	  }
 
-	  a[(j + j * m) * 2 + 0] = ((double) rand() / (double) RAND_MAX) + 8.;
-	  a[(j + j * m) * 2 + 1] = 0.;
+	  a[((long)j + (long)j * (long)m) * 2 + 0] = ((double) rand() / (double) RAND_MAX) + 8.;
+	  a[((long)j + (long)j * (long)m) * 2 + 1] = 0.;
 
 	  for(i = j + 1; i < m; i++) {
-	    a[(i + j * m) * 2 + 0] = 0.;
-	    a[(i + j * m) * 2 + 1] = 0.;
+	    a[((long)i + (long)j * (long)m) * 2 + 0] = 0.;
+	    a[((long)i + (long)j * (long)m) * 2 + 1] = 0.;
 	  }
 	}
       }
diff --git a/benchmark/rot.c b/benchmark/rot.c
index 3ff783cc6..8ec8b1d97 100644
--- a/benchmark/rot.c
+++ b/benchmark/rot.c
@@ -32,9 +32,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #endif
 #include "common.h"
 
+#undef ROT
 
-#undef DOT
-
+#ifndef COMPLEX
 
 #ifdef DOUBLE
 #define ROT   BLASFUNC(drot)
@@ -42,6 +42,15 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define ROT   BLASFUNC(srot)
 #endif
 
+#else
+
+#ifdef DOUBLE
+#define ROT   BLASFUNC(zdrot)
+#else
+#define ROT   BLASFUNC(csrot)
+#endif
+
+#endif
 
 #if defined(__WIN32__) || defined(__WIN64__)
 
@@ -160,17 +169,16 @@ int main(int argc, char *argv[]){
 
    fprintf(stderr, " %6d : ", (int)m);
 
+   for(i = 0; i < m * COMPSIZE * abs(inc_x); i++){
+		x[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+   }
+
+   for(i = 0; i < m * COMPSIZE * abs(inc_y); i++){
+		y[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+   }
 
    for (l=0; l<loops; l++)
    {
-
-   	for(i = 0; i < m * COMPSIZE * abs(inc_x); i++){
-			x[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
-   	}
-
-   	for(i = 0; i < m * COMPSIZE * abs(inc_y); i++){
-			y[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
-   	}
     	gettimeofday( &start, (struct timezone *)0);
 
     	ROT (&m, x, &inc_x, y, &inc_y, c, s);
@@ -179,13 +187,13 @@ int main(int argc, char *argv[]){
 
     	time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
 
-	timeg += time1;
+	    timeg += time1;
 
-    }
+   }
 
-    timeg /= loops;
+   timeg /= loops;
 
-    fprintf(stderr,
+   fprintf(stderr,
 	    " %10.2f MFlops %10.6f sec\n",
 	    COMPSIZE * COMPSIZE * 6. * (double)m / timeg * 1.e-6, timeg);
 
diff --git a/benchmark/rotm.c b/benchmark/rotm.c
new file mode 100644
index 000000000..8dea2d08c
--- /dev/null
+++ b/benchmark/rotm.c
@@ -0,0 +1,210 @@
+/***************************************************************************
+Copyright (c) 2014, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE
+GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF
+THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include <stdio.h>
+#include <stdlib.h>
+#ifdef __CYGWIN32__
+#include <sys/time.h>
+#endif
+#include "common.h"
+
+#undef ROTM
+
+#ifdef DOUBLE
+#define ROTM BLASFUNC(drotm)
+#else
+#define ROTM BLASFUNC(srotm)
+#endif
+
+#if defined(__WIN32__) || defined(__WIN64__)
+
+#ifndef DELTA_EPOCH_IN_MICROSECS
+#define DELTA_EPOCH_IN_MICROSECS 11644473600000000ULL
+#endif
+
+int gettimeofday(struct timeval *tv, void *tz)
+{
+
+    FILETIME ft;
+    unsigned __int64 tmpres = 0;
+    static int tzflag;
+
+    if (NULL != tv) {
+        GetSystemTimeAsFileTime(&ft);
+
+        tmpres |= ft.dwHighDateTime;
+        tmpres <<= 32;
+        tmpres |= ft.dwLowDateTime;
+
+        /*converting file time to unix epoch*/
+        tmpres /= 10; /*convert into microseconds*/
+        tmpres -= DELTA_EPOCH_IN_MICROSECS;
+        tv->tv_sec = (long)(tmpres / 1000000UL);
+        tv->tv_usec = (long)(tmpres % 1000000UL);
+    }
+
+    return 0;
+}
+
+#endif
+
+#if !defined(__WIN32__) && !defined(__WIN64__) && !defined(__CYGWIN32__) && 0
+
+static void *huge_malloc(BLASLONG size)
+{
+    int shmid;
+    void *address;
+
+#ifndef SHM_HUGETLB
+#define SHM_HUGETLB 04000
+#endif
+
+    if ((shmid =
+             shmget(IPC_PRIVATE, (size + HUGE_PAGESIZE) & ~(HUGE_PAGESIZE - 1),
+                    SHM_HUGETLB | IPC_CREAT | 0600)) < 0) {
+        printf("Memory allocation failed(shmget).\n");
+        exit(1);
+    }
+
+    address = shmat(shmid, NULL, SHM_RND);
+
+    if ((BLASLONG)address == -1) {
+        printf("Memory allocation failed(shmat).\n");
+        exit(1);
+    }
+
+    shmctl(shmid, IPC_RMID, 0);
+
+    return address;
+}
+
+#define malloc huge_malloc
+
+#endif
+
+int main(int argc, char *argv[])
+{
+
+    FLOAT *x, *y;
+    // FLOAT result;
+    blasint m, i;
+    blasint inc_x = 1, inc_y = 1;
+    FLOAT param[5] = {1, 2.0, 3.0, 4.0, 5.0};
+    int loops = 1;
+    int l;
+    char *p;
+
+    int from = 1;
+    int to = 200;
+    int step = 1;
+
+    struct timeval start, stop;
+    double time1, timeg;
+
+    argc--;
+    argv++;
+
+    if (argc > 0) {
+        from = atol(*argv);
+        argc--;
+        argv++;
+    }
+    if (argc > 0) {
+        to = MAX(atol(*argv), from);
+        argc--;
+        argv++;
+    }
+    if (argc > 0) {
+        step = atol(*argv);
+        argc--;
+        argv++;
+    }
+
+    if ((p = getenv("OPENBLAS_LOOPS")))
+        loops = atoi(p);
+    if ((p = getenv("OPENBLAS_INCX")))
+        inc_x = atoi(p);
+    if ((p = getenv("OPENBLAS_INCY")))
+        inc_y = atoi(p);
+
+    fprintf(
+        stderr,
+        "From : %3d  To : %3d Step = %3d Inc_x = %d Inc_y = %d Loops = %d\n",
+        from, to, step, inc_x, inc_y, loops);
+
+    if ((x = (FLOAT *)malloc(sizeof(FLOAT) * to * abs(inc_x) * COMPSIZE)) ==
+        NULL) {
+        fprintf(stderr, "Out of Memory!!\n");
+        exit(1);
+    }
+
+    if ((y = (FLOAT *)malloc(sizeof(FLOAT) * to * abs(inc_y) * COMPSIZE)) ==
+        NULL) {
+        fprintf(stderr, "Out of Memory!!\n");
+        exit(1);
+    }
+
+#ifdef linux
+    srandom(getpid());
+#endif
+
+    fprintf(stderr, "   SIZE       Flops\n");
+
+    for (m = from; m <= to; m += step) {
+
+        timeg = 0;
+
+        fprintf(stderr, " %6d : ", (int)m);
+        for (i = 0; i < m * COMPSIZE * abs(inc_x); i++) {
+            x[i] = ((FLOAT)rand() / (FLOAT)RAND_MAX) - 0.5;
+        }
+
+        for (i = 0; i < m * COMPSIZE * abs(inc_y); i++) {
+            y[i] = ((FLOAT)rand() / (FLOAT)RAND_MAX) - 0.5;
+        }
+
+        for (l = 0; l < loops; l++) {
+            gettimeofday(&start, (struct timezone *)0);
+
+            ROTM(&m, x, &inc_x, y, &inc_y, param);
+
+            gettimeofday(&stop, (struct timezone *)0);
+
+            time1 = (double)(stop.tv_sec - start.tv_sec) +
+                    (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
+
+            timeg += time1;
+        }
+
+        timeg /= loops;
+
+        fprintf(stderr, " %10.2f MFlops %10.6f sec\n",
+                COMPSIZE * COMPSIZE * 6. * (double)m / timeg * 1.e-6, timeg);
+    }
+
+    return 0;
+}
diff --git a/benchmark/spmv.c b/benchmark/spmv.c
new file mode 100644
index 000000000..2a26c9416
--- /dev/null
+++ b/benchmark/spmv.c
@@ -0,0 +1,219 @@
+/***************************************************************************
+Copyright (c) 2014, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include <stdio.h>
+#include <stdlib.h>
+#ifdef __CYGWIN32__
+#include <sys/time.h>
+#endif
+#include "common.h"
+
+
+#undef SPMV
+
+
+#ifndef COMPLEX
+
+#ifdef DOUBLE
+#define SPMV   BLASFUNC(dspmv)
+#else
+#define SPMV   BLASFUNC(sspmv)
+#endif
+
+#else
+
+#ifdef DOUBLE
+#define SPMV   BLASFUNC(zspmv)
+#else
+#define SPMV   BLASFUNC(cspmv)
+#endif
+
+#endif
+
+#if defined(__WIN32__) || defined(__WIN64__)
+
+#ifndef DELTA_EPOCH_IN_MICROSECS
+#define DELTA_EPOCH_IN_MICROSECS 11644473600000000ULL
+#endif
+
+int gettimeofday(struct timeval *tv, void *tz){
+
+  FILETIME ft;
+  unsigned __int64 tmpres = 0;
+  static int tzflag;
+
+  if (NULL != tv)
+    {
+      GetSystemTimeAsFileTime(&ft);
+
+      tmpres |= ft.dwHighDateTime;
+      tmpres <<= 32;
+      tmpres |= ft.dwLowDateTime;
+
+      /*converting file time to unix epoch*/
+      tmpres /= 10;  /*convert into microseconds*/
+      tmpres -= DELTA_EPOCH_IN_MICROSECS;
+      tv->tv_sec = (long)(tmpres / 1000000UL);
+      tv->tv_usec = (long)(tmpres % 1000000UL);
+    }
+
+  return 0;
+}
+
+#endif
+
+#if !defined(__WIN32__) && !defined(__WIN64__) && !defined(__CYGWIN32__) && 0
+
+static void *huge_malloc(BLASLONG size){
+  int shmid;
+  void *address;
+
+#ifndef SHM_HUGETLB
+#define SHM_HUGETLB 04000
+#endif
+
+  if ((shmid =shmget(IPC_PRIVATE,
+		     (size + HUGE_PAGESIZE) & ~(HUGE_PAGESIZE - 1),
+		     SHM_HUGETLB | IPC_CREAT |0600)) < 0) {
+    printf( "Memory allocation failed(shmget).\n");
+    exit(1);
+  }
+
+  address = shmat(shmid, NULL, SHM_RND);
+
+  if ((BLASLONG)address == -1){
+    printf( "Memory allocation failed(shmat).\n");
+    exit(1);
+  }
+
+  shmctl(shmid, IPC_RMID, 0);
+
+  return address;
+}
+
+#define malloc huge_malloc
+
+#endif
+
+int main(int argc, char *argv[]){
+
+  FLOAT *a, *x, *y;
+  FLOAT alpha[] = {1.0, 1.0};
+  FLOAT beta [] = {1.0, 1.0};
+  char uplo='L';
+  blasint m, i, j;
+  blasint inc_x=1,inc_y=1;
+  int loops = 1;
+  int l;
+  char *p;
+
+  int from =   1;
+  int to   = 200;
+  int step =   1;
+
+  struct timeval start, stop;
+  double time1,timeg;
+
+  argc--;argv++;
+
+  if (argc > 0) { from     = atol(*argv);		argc--; argv++;}
+  if (argc > 0) { to       = MAX(atol(*argv), from);	argc--; argv++;}
+  if (argc > 0) { step     = atol(*argv);		argc--; argv++;}
+
+  if ((p = getenv("OPENBLAS_LOOPS")))  loops = atoi(p);
+  if ((p = getenv("OPENBLAS_INCX")))   inc_x = atoi(p);
+  if ((p = getenv("OPENBLAS_INCY")))   inc_y = atoi(p);
+  if ((p = getenv("OPENBLAS_UPLO")))  uplo=*p;
+
+  fprintf(stderr, "From : %3d  To : %3d Step = %3d Uplo = '%c' Inc_x = %d Inc_y = %d Loops = %d\n", from, to, step,uplo,inc_x,inc_y,loops);
+
+  if (( a = (FLOAT *)malloc(sizeof(FLOAT) * to * to * COMPSIZE)) == NULL){
+    fprintf(stderr,"Out of Memory!!\n");exit(1);
+  }
+
+  if (( x = (FLOAT *)malloc(sizeof(FLOAT) * to * abs(inc_x) * COMPSIZE)) == NULL){
+    fprintf(stderr,"Out of Memory!!\n");exit(1);
+  }
+
+  if (( y = (FLOAT *)malloc(sizeof(FLOAT) * to * abs(inc_y) * COMPSIZE)) == NULL){
+    fprintf(stderr,"Out of Memory!!\n");exit(1);
+  }
+
+#ifdef linux
+  srandom(getpid());
+#endif
+
+  fprintf(stderr, "   SIZE       Flops\n");
+
+  for(m = from; m <= to; m += step)
+  {
+
+   timeg=0;
+
+   fprintf(stderr, " %6dx%d : ", (int)m,(int)m);
+
+   for(j = 0; j < m; j++){
+      		for(i = 0; i < m * COMPSIZE; i++){
+			a[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+      		}
+   }
+
+
+    for (l=0; l<loops; l++)
+    {
+
+   	for(i = 0; i < m * COMPSIZE * abs(inc_x); i++){
+			x[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+   	}
+
+   	for(i = 0; i < m * COMPSIZE * abs(inc_y); i++){
+			y[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+   	}
+    	gettimeofday( &start, (struct timezone *)0);
+
+    	SPMV (&uplo, &m, alpha, a, x, &inc_x, beta, y, &inc_y );
+
+    	gettimeofday( &stop, (struct timezone *)0);
+
+    	time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
+
+	timeg += time1;
+
+    }
+
+    timeg /= loops;
+
+    fprintf(stderr,
+	    " %10.2f MFlops\n",
+	    COMPSIZE * COMPSIZE * 2. * (double)m * (double)m / timeg * 1.e-6);
+
+  }
+
+  return 0;
+}
+
+// void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));
diff --git a/benchmark/spr.c b/benchmark/spr.c
new file mode 100755
index 000000000..c91e587b1
--- /dev/null
+++ b/benchmark/spr.c
@@ -0,0 +1,198 @@
+/***************************************************************************
+Copyright (c) 2014, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include <stdio.h>
+#include <stdlib.h>
+#ifdef __CYGWIN32__
+#include <sys/time.h>
+#endif
+#include "common.h"
+
+
+#undef SPR
+
+#ifdef DOUBLE
+#define SPR   BLASFUNC(dspr)
+#else
+#define SPR   BLASFUNC(sspr)
+#endif
+
+
+
+#if defined(__WIN32__) || defined(__WIN64__)
+
+#ifndef DELTA_EPOCH_IN_MICROSECS
+#define DELTA_EPOCH_IN_MICROSECS 11644473600000000ULL
+#endif
+
+int gettimeofday(struct timeval *tv, void *tz){
+
+  FILETIME ft;
+  unsigned __int64 tmpres = 0;
+  static int tzflag;
+
+  if (NULL != tv)
+    {
+      GetSystemTimeAsFileTime(&ft);
+
+      tmpres |= ft.dwHighDateTime;
+      tmpres <<= 32;
+      tmpres |= ft.dwLowDateTime;
+
+      /*converting file time to unix epoch*/
+      tmpres /= 10;  /*convert into microseconds*/
+      tmpres -= DELTA_EPOCH_IN_MICROSECS;
+      tv->tv_sec = (long)(tmpres / 1000000UL);
+      tv->tv_usec = (long)(tmpres % 1000000UL);
+    }
+
+  return 0;
+}
+
+#endif
+
+#if !defined(__WIN32__) && !defined(__WIN64__) && !defined(__CYGWIN32__) && 0
+
+static void *huge_malloc(BLASLONG size){
+  int shmid;
+  void *address;
+
+#ifndef SHM_HUGETLB
+#define SHM_HUGETLB 04000
+#endif
+
+  if ((shmid =shmget(IPC_PRIVATE,
+		     (size + HUGE_PAGESIZE) & ~(HUGE_PAGESIZE - 1),
+		     SHM_HUGETLB | IPC_CREAT |0600)) < 0) {
+    printf( "Memory allocation failed(shmget).\n");
+    exit(1);
+  }
+
+  address = shmat(shmid, NULL, SHM_RND);
+
+  if ((BLASLONG)address == -1){
+    printf( "Memory allocation failed(shmat).\n");
+    exit(1);
+  }
+
+  shmctl(shmid, IPC_RMID, 0);
+
+  return address;
+}
+
+#define malloc huge_malloc
+
+#endif
+
+int main(int argc, char *argv[]){
+
+  FLOAT *a,*c;
+  FLOAT alpha[] = {1.0, 1.0};
+  blasint inc_x=1;
+  int loops = 1;
+  int l;
+  char *p;
+
+  char uplo='U';
+
+  if ((p = getenv("OPENBLAS_UPLO"))) uplo=*p;
+  if ((p = getenv("OPENBLAS_LOOPS")))  loops = atoi(p);
+  if ((p = getenv("OPENBLAS_INCX")))   inc_x = atoi(p);
+
+  blasint m, i, j;
+
+  int from =   1;
+  int to   = 200;
+  int step =   1;
+
+  struct timeval start, stop;
+  double time1,timeg;
+
+  argc--;argv++;
+
+  if (argc > 0) { from     = atol(*argv);		argc--; argv++;}
+  if (argc > 0) { to       = MAX(atol(*argv), from);	argc--; argv++;}
+  if (argc > 0) { step     = atol(*argv);		argc--; argv++;}
+
+  fprintf(stderr, "From : %3d  To : %3d Step = %3d Uplo = %c Inc_x = %d\n", from, to, step,uplo,inc_x);
+
+
+  if (( a = (FLOAT *)malloc(sizeof(FLOAT) * to * to * COMPSIZE)) == NULL){
+    fprintf(stderr,"Out of Memory!!\n");exit(1);
+  }
+
+  if (( c = (FLOAT *)malloc(sizeof(FLOAT) * to * abs(inc_x) * COMPSIZE)) == NULL){
+    fprintf(stderr,"Out of Memory!!\n");exit(1);
+  }
+
+#ifdef linux
+  srandom(getpid());
+#endif
+
+  fprintf(stderr, "   SIZE       Flops		Time\n");
+
+  for(m = from; m <= to; m += step)
+  {
+	timeg=0;
+	
+    fprintf(stderr, " %6d : ", (int)m);
+	
+	for (l=0; l<loops; l++)
+    {
+
+		for(j = 0; j < m; j++){
+		  for(i = 0; i < m * COMPSIZE; i++){
+			a[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+		  }
+		}
+		for(i = 0; i < m * COMPSIZE * abs(inc_x); i++){
+			c[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+		}
+
+		gettimeofday( &start, (struct timezone *)0);
+
+		SPR (&uplo, &m, alpha, c, &inc_x, a);
+
+		gettimeofday( &stop, (struct timezone *)0);
+
+		time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
+		
+		timeg += time1;
+   }
+	
+	timeg /= loops;
+	 
+    fprintf(stderr,
+	    " %10.2f MFlops %10.6f sec\n",
+	    COMPSIZE * COMPSIZE * 1. * (double)m * (double)m  / timeg * 1.e-6, timeg);
+
+  }
+
+  return 0;
+}
+
+// void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));
diff --git a/benchmark/spr2.c b/benchmark/spr2.c
new file mode 100755
index 000000000..e8ee345d7
--- /dev/null
+++ b/benchmark/spr2.c
@@ -0,0 +1,207 @@
+/***************************************************************************
+Copyright (c) 2014, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include <stdio.h>
+#include <stdlib.h>
+#ifdef __CYGWIN32__
+#include <sys/time.h>
+#endif
+#include "common.h"
+
+
+#undef SPR2
+
+#ifdef DOUBLE
+#define SPR2   BLASFUNC(dspr2)
+#else
+#define SPR2   BLASFUNC(sspr2)
+#endif
+
+
+
+#if defined(__WIN32__) || defined(__WIN64__)
+
+#ifndef DELTA_EPOCH_IN_MICROSECS
+#define DELTA_EPOCH_IN_MICROSECS 11644473600000000ULL
+#endif
+
+int gettimeofday(struct timeval *tv, void *tz){
+
+  FILETIME ft;
+  unsigned __int64 tmpres = 0;
+  static int tzflag;
+
+  if (NULL != tv)
+    {
+      GetSystemTimeAsFileTime(&ft);
+
+      tmpres |= ft.dwHighDateTime;
+      tmpres <<= 32;
+      tmpres |= ft.dwLowDateTime;
+
+      /*converting file time to unix epoch*/
+      tmpres /= 10;  /*convert into microseconds*/
+      tmpres -= DELTA_EPOCH_IN_MICROSECS;
+      tv->tv_sec = (long)(tmpres / 1000000UL);
+      tv->tv_usec = (long)(tmpres % 1000000UL);
+    }
+
+  return 0;
+}
+
+#endif
+
+#if !defined(__WIN32__) && !defined(__WIN64__) && !defined(__CYGWIN32__) && 0
+
+static void *huge_malloc(BLASLONG size){
+  int shmid;
+  void *address;
+
+#ifndef SHM_HUGETLB
+#define SHM_HUGETLB 04000
+#endif
+
+  if ((shmid =shmget(IPC_PRIVATE,
+		     (size + HUGE_PAGESIZE) & ~(HUGE_PAGESIZE - 1),
+		     SHM_HUGETLB | IPC_CREAT |0600)) < 0) {
+    printf( "Memory allocation failed(shmget).\n");
+    exit(1);
+  }
+
+  address = shmat(shmid, NULL, SHM_RND);
+
+  if ((BLASLONG)address == -1){
+    printf( "Memory allocation failed(shmat).\n");
+    exit(1);
+  }
+
+  shmctl(shmid, IPC_RMID, 0);
+
+  return address;
+}
+
+#define malloc huge_malloc
+
+#endif
+
+int main(int argc, char *argv[]){
+
+  FLOAT *a,*b,*c;
+  FLOAT alpha[] = {1.0, 1.0};
+  blasint inc_x=1,inc_y=1;
+  int loops = 1;
+  int l;
+  char *p;
+
+  char uplo='U';
+
+  if ((p = getenv("OPENBLAS_UPLO"))) uplo=*p;
+  if ((p = getenv("OPENBLAS_LOOPS")))  loops = atoi(p);
+  if ((p = getenv("OPENBLAS_INCX")))   inc_x = atoi(p);
+
+  blasint m, i, j;
+
+  int from =   1;
+  int to   = 200;
+  int step =   1;
+
+  struct timeval start, stop;
+  double time1,timeg;
+
+  argc--;argv++;
+
+  if (argc > 0) { from     = atol(*argv);		argc--; argv++;}
+  if (argc > 0) { to       = MAX(atol(*argv), from);	argc--; argv++;}
+  if (argc > 0) { step     = atol(*argv);		argc--; argv++;}
+
+  fprintf(stderr, "From : %3d  To : %3d Step = %3d Uplo = %c Inc_x = %d Inc_y = %d\n", from, to, step,uplo,inc_x,inc_y);
+
+
+  if (( a = (FLOAT *)malloc(sizeof(FLOAT) * to * to * COMPSIZE)) == NULL){
+    fprintf(stderr,"Out of Memory!!\n");exit(1);
+  }
+  
+  if (( b = (FLOAT *)malloc(sizeof(FLOAT) * to * abs(inc_y) * COMPSIZE)) == NULL){
+    fprintf(stderr,"Out of Memory!!\n");exit(1);
+  }
+
+  if (( c = (FLOAT *)malloc(sizeof(FLOAT) * to * abs(inc_x) * COMPSIZE)) == NULL){
+    fprintf(stderr,"Out of Memory!!\n");exit(1);
+  }
+
+#ifdef linux
+  srandom(getpid());
+#endif
+
+  fprintf(stderr, "   SIZE       Flops		Time\n");
+
+  for(m = from; m <= to; m += step)
+  {
+    timeg=0;
+	
+    fprintf(stderr, " %6d : ", (int)m);
+	
+	for (l=0; l<loops; l++)
+    {
+
+		for(j = 0; j < m; j++){
+		  for(i = 0; i < m * COMPSIZE; i++){
+			a[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+		  }
+		}
+		
+		for(i = 0; i < m * COMPSIZE * abs(inc_y); i++){
+			b[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+		}
+		
+		for(i = 0; i < m * COMPSIZE * abs(inc_x); i++){
+			c[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+		}
+
+		gettimeofday( &start, (struct timezone *)0);
+
+		SPR2 (&uplo, &m, alpha, c, &inc_x, b, &inc_y, a);
+
+		gettimeofday( &stop, (struct timezone *)0);
+
+		time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
+		
+		timeg += time1;
+   }
+	
+    timeg /= loops;
+	 
+    fprintf(stderr,
+	    " %10.2f MFlops %10.6f sec\n",
+	    COMPSIZE * COMPSIZE * 2. * (double)m * (double)m  / timeg * 1.e-6, timeg);
+
+  }
+
+  return 0;
+}
+
+// void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));
diff --git a/benchmark/symm.c b/benchmark/symm.c
index 35ebcee97..b979e8d51 100644
--- a/benchmark/symm.c
+++ b/benchmark/symm.c
@@ -175,9 +175,9 @@ int main(int argc, char *argv[]){
 
     for(j = 0; j < m; j++){
       for(i = 0; i < m * COMPSIZE; i++){
-	a[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
-	b[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
-	c[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+	a[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+	b[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+	c[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
       }
     }
 
@@ -189,8 +189,6 @@ int main(int argc, char *argv[]){
 
     time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
 
-    gettimeofday( &start, (struct timezone *)0);
-
     fprintf(stderr,
 	    " %10.2f MFlops\n",
 	    COMPSIZE * COMPSIZE * 2. * (double)m * (double)m * (double)m / time1 * 1.e-6);
diff --git a/benchmark/symv.c b/benchmark/symv.c
index df2a5d301..789c3560f 100644
--- a/benchmark/symv.c
+++ b/benchmark/symv.c
@@ -177,7 +177,7 @@ int main(int argc, char *argv[]){
 
    for(j = 0; j < m; j++){
       		for(i = 0; i < m * COMPSIZE; i++){
-			a[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+			a[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
       		}
    }
 
diff --git a/benchmark/syr.c b/benchmark/syr.c
index 91b5b5904..458bc6edb 100644
--- a/benchmark/syr.c
+++ b/benchmark/syr.c
@@ -173,11 +173,9 @@ int main(int argc, char *argv[]){
 
     time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
 
-    gettimeofday( &start, (struct timezone *)0);
-
     fprintf(stderr,
 	    " %10.2f MFlops\n",
-	    COMPSIZE * COMPSIZE * 1. * (double)m * (double)m * (double)m / time1 * 1.e-6);
+	    COMPSIZE * COMPSIZE * 1. * (double)m * (double)m / time1 * 1.e-6);
 
   }
 
diff --git a/benchmark/syr2.c b/benchmark/syr2.c
new file mode 100644
index 000000000..0129dd09a
--- /dev/null
+++ b/benchmark/syr2.c
@@ -0,0 +1,194 @@
+/***************************************************************************
+Copyright (c) 2014, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include <stdio.h>
+#include <stdlib.h>
+#ifdef __CYGWIN32__
+#include <sys/time.h>
+#endif
+#include "common.h"
+
+
+#undef SYR2
+
+
+#ifdef DOUBLE
+#define SYR2   BLASFUNC(dsyr2)
+#else
+#define SYR2   BLASFUNC(ssyr2)
+#endif
+
+
+#if defined(__WIN32__) || defined(__WIN64__)
+
+#ifndef DELTA_EPOCH_IN_MICROSECS
+#define DELTA_EPOCH_IN_MICROSECS 11644473600000000ULL
+#endif
+
+int gettimeofday(struct timeval *tv, void *tz){
+
+  FILETIME ft;
+  unsigned __int64 tmpres = 0;
+  static int tzflag;
+
+  if (NULL != tv)
+    {
+      GetSystemTimeAsFileTime(&ft);
+
+      tmpres |= ft.dwHighDateTime;
+      tmpres <<= 32;
+      tmpres |= ft.dwLowDateTime;
+
+      /*converting file time to unix epoch*/
+      tmpres /= 10;  /*convert into microseconds*/
+      tmpres -= DELTA_EPOCH_IN_MICROSECS;
+      tv->tv_sec = (long)(tmpres / 1000000UL);
+      tv->tv_usec = (long)(tmpres % 1000000UL);
+    }
+
+  return 0;
+}
+
+#endif
+
+#if !defined(__WIN32__) && !defined(__WIN64__) && !defined(__CYGWIN32__) && 0
+
+static void *huge_malloc(BLASLONG size){
+  int shmid;
+  void *address;
+
+#ifndef SHM_HUGETLB
+#define SHM_HUGETLB 04000
+#endif
+
+  if ((shmid =shmget(IPC_PRIVATE,
+		     (size + HUGE_PAGESIZE) & ~(HUGE_PAGESIZE - 1),
+		     SHM_HUGETLB | IPC_CREAT |0600)) < 0) {
+    printf( "Memory allocation failed(shmget).\n");
+    exit(1);
+  }
+
+  address = shmat(shmid, NULL, SHM_RND);
+
+  if ((BLASLONG)address == -1){
+    printf( "Memory allocation failed(shmat).\n");
+    exit(1);
+  }
+
+  shmctl(shmid, IPC_RMID, 0);
+
+  return address;
+}
+
+#define malloc huge_malloc
+
+#endif
+
+int main(int argc, char *argv[]){
+
+  FLOAT *x, *y, *a;
+  FLOAT alpha[] = {1.0, 1.0};
+  char *p;
+
+  char uplo='U';
+
+  if ((p = getenv("OPENBLAS_UPLO"))) uplo=*p; 
+
+  blasint m, i, j;
+  blasint inc_x= 1;
+  blasint inc_y= 1;
+  int from =   1;
+  int to   = 200;
+  int step =   1;
+
+  struct timeval start, stop;
+  double time1;
+
+  argc--;argv++;
+
+  if (argc > 0) { from     = atol(*argv);		argc--; argv++;}
+  if (argc > 0) { to       = MAX(atol(*argv), from);	argc--; argv++;}
+  if (argc > 0) { step     = atol(*argv);		argc--; argv++;}
+
+  fprintf(stderr, "From : %3d  To : %3d Step = %3d Uplo = %c Inc_x = %d Inc_y = %d\n", from, to, step,uplo,inc_x,inc_y);
+
+  if (( a = (FLOAT *)malloc(sizeof(FLOAT) * to * to * COMPSIZE)) == NULL){
+    fprintf(stderr,"Out of Memory!!\n");exit(1);
+  }
+
+   if (( x = (FLOAT *)malloc(sizeof(FLOAT) * to * abs(inc_x) * COMPSIZE)) == NULL){
+    fprintf(stderr,"Out of Memory!!\n");exit(1);
+  }
+  
+   if (( y = (FLOAT *)malloc(sizeof(FLOAT) * to * abs(inc_y) * COMPSIZE)) == NULL){
+    fprintf(stderr,"Out of Memory!!\n");exit(1);
+  }
+
+
+
+#ifdef linux
+  srandom(getpid());
+#endif
+
+  fprintf(stderr, "   SIZE       Flops\n");
+
+  for(m = from; m <= to; m += step)
+  {
+	
+    fprintf(stderr, " %6d : ", (int)m);
+	for(i = 0; i < m * COMPSIZE * abs(inc_x); i++){
+	x[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+   	}
+	
+	for(i = 0; i < m * COMPSIZE * abs(inc_y); i++){
+	y[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+   	}
+
+    for(j = 0; j < m; j++){
+      for(i = 0; i < m * COMPSIZE; i++){
+	a[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+      }
+    }
+
+    gettimeofday( &start, (struct timezone *)0);
+
+    SYR2 (&uplo, &m, alpha, x, &inc_x, y, &inc_y, a, &m );
+
+    gettimeofday( &stop, (struct timezone *)0);
+
+    time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
+
+    fprintf(stderr,
+	    " %10.2f MFlops\n",
+	    COMPSIZE * COMPSIZE * 2. * (double)m * (double)m / time1 * 1.e-6);
+
+  }
+
+  return 0;
+}
+
+// void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));
diff --git a/benchmark/syr2k.c b/benchmark/syr2k.c
index 9840b5f3e..b1fcd8a18 100644
--- a/benchmark/syr2k.c
+++ b/benchmark/syr2k.c
@@ -175,9 +175,9 @@ int main(int argc, char *argv[]){
 
     for(j = 0; j < m; j++){
       for(i = 0; i < m * COMPSIZE; i++){
-	a[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
-	b[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
-	c[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+	a[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+	b[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+	c[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
       }
     }
 
@@ -189,8 +189,6 @@ int main(int argc, char *argv[]){
 
     time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
 
-    gettimeofday( &start, (struct timezone *)0);
-
     fprintf(stderr,
 	    " %10.2f MFlops\n",
 	    COMPSIZE * COMPSIZE * 2. * (double)m * (double)m * (double)m / time1 * 1.e-6);
diff --git a/benchmark/syrk.c b/benchmark/syrk.c
index 34817f2bb..95625a6c4 100644
--- a/benchmark/syrk.c
+++ b/benchmark/syrk.c
@@ -172,8 +172,8 @@ int main(int argc, char *argv[]){
 
     for(j = 0; j < m; j++){
       for(i = 0; i < m * COMPSIZE; i++){
-	a[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
-	c[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+	a[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+	c[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
       }
     }
 
@@ -185,8 +185,6 @@ int main(int argc, char *argv[]){
 
     time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
 
-    gettimeofday( &start, (struct timezone *)0);
-
     fprintf(stderr,
 	    " %10.2f MFlops\n",
 	    COMPSIZE * COMPSIZE * 1. * (double)m * (double)m * (double)m / time1 * 1.e-6);
diff --git a/benchmark/tpmv.c b/benchmark/tpmv.c
new file mode 100644
index 000000000..ee5b97f24
--- /dev/null
+++ b/benchmark/tpmv.c
@@ -0,0 +1,172 @@
+/***************************************************************************
+Copyright (c) 2014, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include <stdio.h>
+#include <stdlib.h>
+#ifdef __CYGWIN32__
+#include <sys/time.h>
+#endif
+#include "common.h"
+
+#undef TPMV
+
+#ifndef COMPLEX
+
+#ifdef DOUBLE
+#define TPMV   BLASFUNC(dtpmv)
+#else
+#define TPMV   BLASFUNC(stpmv)
+#endif
+
+#else
+
+#ifdef DOUBLE
+#define TPMV   BLASFUNC(ztpmv)
+#else
+#define TPMV   BLASFUNC(ctpmv)
+#endif
+
+#endif
+
+#if !defined(__WIN32__) && !defined(__WIN64__) && !defined(__CYGWIN32__) && 0
+
+static void *huge_malloc(BLASLONG size)
+{
+    int shmid;
+    void *address;
+
+#ifndef SHM_HUGETLB
+#define SHM_HUGETLB 04000
+#endif
+
+    if ((shmid =shmget(IPC_PRIVATE,
+             (size + HUGE_PAGESIZE) & ~(HUGE_PAGESIZE - 1),
+             SHM_HUGETLB | IPC_CREAT |0600)) < 0) {
+        printf( "Memory allocation failed(shmget).\n");
+        exit(1);
+    }
+
+    address = shmat(shmid, NULL, SHM_RND);
+
+    if ((BLASLONG)address == -1) {
+        printf( "Memory allocation failed(shmat).\n");
+        exit(1);
+    }
+
+    shmctl(shmid, IPC_RMID, 0);
+
+    return address;
+}
+
+#define malloc huge_malloc
+
+#endif
+
+int main(int argc, char *argv[])
+{
+
+    FLOAT *a, *x;
+    char *p;
+
+    char uplo ='U';
+    char trans='N';
+    char diag ='U';
+
+    int loops = 1;
+    int l;
+    blasint inc_x=1;
+
+    if ((p = getenv("OPENBLAS_UPLO"))) uplo=*p;
+    if ((p = getenv("OPENBLAS_TRANS"))) trans=*p;
+    if ((p = getenv("OPENBLAS_DIAG"))) diag=*p;
+    if ((p = getenv("OPENBLAS_LOOPS")))  loops = atoi(p);
+    if ((p = getenv("OPENBLAS_INCX")))   inc_x = atoi(p);
+
+    blasint n, i, j;
+
+    int from =   1;
+    int to   = 200;
+    int step =   1;
+
+    struct timespec start = { 0, 0 }, stop = { 0, 0 };
+    double time1, timeg;
+
+    argc--;argv++;
+
+    if (argc > 0) { from     = atol(*argv);        argc--; argv++;}
+    if (argc > 0) { to       = MAX(atol(*argv), from);    argc--; argv++;}
+    if (argc > 0) { step     = atol(*argv);        argc--; argv++;}
+
+    fprintf(stderr, "From : %3d  To : %3d Step = %3d Uplo = %c Trans = %c  Diag = %c Loops=%d Inc_x=%d\n", from,
+            to, step, uplo, trans, diag, loops, inc_x);
+
+    if (( a = (FLOAT *)malloc(sizeof(FLOAT) * to * to * COMPSIZE)) == NULL) {
+        fprintf(stderr,"Out of Memory!!\n");exit(1);
+    }
+
+    if (( x = (FLOAT *)malloc(sizeof(FLOAT) * to * abs(inc_x) * COMPSIZE)) == NULL){
+        fprintf(stderr,"Out of Memory!!\n");exit(1);
+    }
+
+#ifdef linux
+    srandom(getpid());
+#endif
+
+    fprintf(stderr, "   SIZE       Flops\n");
+
+    for(n = from; n <= to; n += step) {
+        timeg=0;
+
+        fprintf(stderr, " %6d : ", (int)n);
+        for(j = 0; j < n; j++) {
+            for(i = 0; i < n * COMPSIZE; i++) {
+                a[(long)i + (long)j * (long)n * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+            }
+        }
+
+        for (i = 0; i < n * COMPSIZE * abs(inc_x); i++) {
+            x[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+        }
+
+        for (l = 0; l < loops; l++) {
+            clock_gettime(CLOCK_REALTIME, &start);
+            TPMV (&uplo, &trans, &diag, &n, a, x, &inc_x);
+            clock_gettime(CLOCK_REALTIME, &stop);
+
+            time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_nsec - start.tv_nsec)) / 1.e9;
+            timeg += time1;
+        }
+
+        timeg /= loops;
+        fprintf(stderr, " %10.2f MFlops %12.9f sec\n",
+                COMPSIZE * COMPSIZE * 1. * (double)n * (double)n / timeg / 1.e6, timeg);
+    }
+
+    return 0;
+}
+
+// void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));
diff --git a/benchmark/tpsv.c b/benchmark/tpsv.c
new file mode 100644
index 000000000..46d78fd17
--- /dev/null
+++ b/benchmark/tpsv.c
@@ -0,0 +1,172 @@
+/***************************************************************************
+Copyright (c) 2014, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include <stdio.h>
+#include <stdlib.h>
+#ifdef __CYGWIN32__
+#include <sys/time.h>
+#endif
+#include "common.h"
+
+#undef TPSV
+
+#ifndef COMPLEX
+
+#ifdef DOUBLE
+#define TPSV   BLASFUNC(dtpsv)
+#else
+#define TPSV   BLASFUNC(stpsv)
+#endif
+
+#else
+
+#ifdef DOUBLE
+#define TPSV   BLASFUNC(ztpsv)
+#else
+#define TPSV   BLASFUNC(ctpsv)
+#endif
+
+#endif
+
+#if !defined(__WIN32__) && !defined(__WIN64__) && !defined(__CYGWIN32__) && 0
+
+static void *huge_malloc(BLASLONG size)
+{
+    int shmid;
+    void *address;
+
+#ifndef SHM_HUGETLB
+#define SHM_HUGETLB 04000
+#endif
+
+    if ((shmid =shmget(IPC_PRIVATE,
+             (size + HUGE_PAGESIZE) & ~(HUGE_PAGESIZE - 1),
+             SHM_HUGETLB | IPC_CREAT |0600)) < 0) {
+        printf( "Memory allocation failed(shmget).\n");
+        exit(1);
+    }
+
+    address = shmat(shmid, NULL, SHM_RND);
+
+    if ((BLASLONG)address == -1) {
+        printf( "Memory allocation failed(shmat).\n");
+        exit(1);
+    }
+
+    shmctl(shmid, IPC_RMID, 0);
+
+    return address;
+}
+
+#define malloc huge_malloc
+
+#endif
+
+int main(int argc, char *argv[])
+{
+
+    FLOAT *a, *x;
+    char *p;
+
+    char uplo ='U';
+    char trans='N';
+    char diag ='U';
+
+    int loops = 1;
+    int l;
+    blasint inc_x=1;
+
+    if ((p = getenv("OPENBLAS_UPLO"))) uplo=*p;
+    if ((p = getenv("OPENBLAS_TRANS"))) trans=*p;
+    if ((p = getenv("OPENBLAS_DIAG"))) diag=*p;
+    if ((p = getenv("OPENBLAS_LOOPS")))  loops = atoi(p);
+    if ((p = getenv("OPENBLAS_INCX")))   inc_x = atoi(p);
+
+    blasint n, i, j;
+
+    int from =   1;
+    int to   = 200;
+    int step =   1;
+
+    struct timespec start = { 0, 0 }, stop = { 0, 0 };
+    double time1, timeg;
+
+    argc--;argv++;
+
+    if (argc > 0) { from     = atol(*argv);        argc--; argv++;}
+    if (argc > 0) { to       = MAX(atol(*argv), from);    argc--; argv++;}
+    if (argc > 0) { step     = atol(*argv);        argc--; argv++;}
+
+    fprintf(stderr, "From : %3d  To : %3d Step = %3d Uplo = %c Trans = %c  Diag = %c Loops=%d Inc_x=%d\n", from,
+            to, step, uplo, trans, diag, loops, inc_x);
+
+    if (( a = (FLOAT *)malloc(sizeof(FLOAT) * to * to * COMPSIZE)) == NULL) {
+        fprintf(stderr,"Out of Memory!!\n");exit(1);
+    }
+
+    if (( x = (FLOAT *)malloc(sizeof(FLOAT) * to * abs(inc_x) * COMPSIZE)) == NULL){
+        fprintf(stderr,"Out of Memory!!\n");exit(1);
+    }
+
+#ifdef linux
+    srandom(getpid());
+#endif
+
+    fprintf(stderr, "   SIZE       Flops\n");
+
+    for(n = from; n <= to; n += step) {
+        timeg=0;
+
+        fprintf(stderr, " %6d : ", (int)n);
+        for(j = 0; j < n; j++) {
+            for(i = 0; i < n * COMPSIZE; i++) {
+                a[(long)i + (long)j * (long)n * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+            }
+        }
+
+        for (i = 0; i < n * COMPSIZE * abs(inc_x); i++) {
+            x[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+        }
+
+        for (l = 0; l < loops; l++) {
+            clock_gettime(CLOCK_REALTIME, &start);
+            TPSV (&uplo, &trans, &diag, &n, a, x, &inc_x);
+            clock_gettime(CLOCK_REALTIME, &stop);
+
+            time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_nsec - start.tv_nsec)) / 1.e9;
+            timeg += time1;
+        }
+
+        timeg /= loops;
+        fprintf(stderr, " %10.2f MFlops %12.9f sec\n",
+                COMPSIZE * COMPSIZE * 1. * (double)n * (double)n / timeg / 1.e6, timeg);
+    }
+
+    return 0;
+}
+
+// void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));
diff --git a/benchmark/trmm.c b/benchmark/trmm.c
index 54c7972db..e095b85ee 100644
--- a/benchmark/trmm.c
+++ b/benchmark/trmm.c
@@ -175,8 +175,8 @@ int main(int argc, char *argv[]){
 
     for(j = 0; j < m; j++){
       for(i = 0; i < m * COMPSIZE; i++){
-	a[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
-	b[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+	a[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+	b[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
       }
     }
 
@@ -188,8 +188,6 @@ int main(int argc, char *argv[]){
 
     time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
 
-    gettimeofday( &start, (struct timezone *)0);
-
     fprintf(stderr,
 	    " %10.2f MFlops  %10.6f sec\n",
 	    COMPSIZE * COMPSIZE * 1. * (double)m * (double)m * (double)m / time1 * 1.e-6, time1);
diff --git a/benchmark/trmv.c b/benchmark/trmv.c
new file mode 100644
index 000000000..f5a5fe31a
--- /dev/null
+++ b/benchmark/trmv.c
@@ -0,0 +1,172 @@
+/***************************************************************************
+Copyright (c) 2014, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include <stdio.h>
+#include <stdlib.h>
+#ifdef __CYGWIN32__
+#include <sys/time.h>
+#endif
+#include "common.h"
+
+#undef TRMV
+
+#ifndef COMPLEX
+
+#ifdef DOUBLE
+#define TRMV   BLASFUNC(dtrmv)
+#else
+#define TRMV   BLASFUNC(strmv)
+#endif
+
+#else
+
+#ifdef DOUBLE
+#define TRMV   BLASFUNC(ztrmv)
+#else
+#define TRMV   BLASFUNC(ctrmv)
+#endif
+
+#endif
+
+#if !defined(__WIN32__) && !defined(__WIN64__) && !defined(__CYGWIN32__) && 0
+
+static void *huge_malloc(BLASLONG size)
+{
+    int shmid;
+    void *address;
+
+#ifndef SHM_HUGETLB
+#define SHM_HUGETLB 04000
+#endif
+
+    if ((shmid =shmget(IPC_PRIVATE,
+             (size + HUGE_PAGESIZE) & ~(HUGE_PAGESIZE - 1),
+             SHM_HUGETLB | IPC_CREAT |0600)) < 0) {
+        printf( "Memory allocation failed(shmget).\n");
+        exit(1);
+    }
+
+    address = shmat(shmid, NULL, SHM_RND);
+
+    if ((BLASLONG)address == -1) {
+        printf( "Memory allocation failed(shmat).\n");
+        exit(1);
+    }
+
+    shmctl(shmid, IPC_RMID, 0);
+
+    return address;
+}
+
+#define malloc huge_malloc
+
+#endif
+
+int main(int argc, char *argv[])
+{
+
+    FLOAT *a, *x;
+    char *p;
+
+    char uplo ='U';
+    char trans='N';
+    char diag ='U';
+
+    int loops = 1;
+    int l;
+    blasint inc_x=1;
+
+    if ((p = getenv("OPENBLAS_UPLO"))) uplo=*p;
+    if ((p = getenv("OPENBLAS_TRANS"))) trans=*p;
+    if ((p = getenv("OPENBLAS_DIAG"))) diag=*p;
+    if ((p = getenv("OPENBLAS_LOOPS")))  loops = atoi(p);
+    if ((p = getenv("OPENBLAS_INCX")))   inc_x = atoi(p);
+
+    blasint n, i, j;
+
+    int from =   1;
+    int to   = 200;
+    int step =   1;
+
+    struct timespec start = { 0, 0 }, stop = { 0, 0 };
+    double time1, timeg;
+
+    argc--;argv++;
+
+    if (argc > 0) { from     = atol(*argv);        argc--; argv++;}
+    if (argc > 0) { to       = MAX(atol(*argv), from);    argc--; argv++;}
+    if (argc > 0) { step     = atol(*argv);        argc--; argv++;}
+
+    fprintf(stderr, "From : %3d  To : %3d Step = %3d Uplo = %c Trans = %c  Diag = %c Loops=%d Inc_x=%d\n", from,
+            to, step, uplo, trans, diag, loops, inc_x);
+
+    if (( a = (FLOAT *)malloc(sizeof(FLOAT) * to * to * COMPSIZE)) == NULL) {
+        fprintf(stderr,"Out of Memory!!\n");exit(1);
+    }
+
+    if (( x = (FLOAT *)malloc(sizeof(FLOAT) * to * abs(inc_x) * COMPSIZE)) == NULL){
+        fprintf(stderr,"Out of Memory!!\n");exit(1);
+    }
+
+#ifdef linux
+    srandom(getpid());
+#endif
+
+    fprintf(stderr, "   SIZE       Flops\n");
+
+    for(n = from; n <= to; n += step) {
+        timeg=0;
+
+        fprintf(stderr, " %6d : ", (int)n);
+        for(j = 0; j < n; j++) {
+            for(i = 0; i < n * COMPSIZE; i++) {
+                a[(long)i + (long)j * (long)n * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+            }
+        }
+
+        for (i = 0; i < n * COMPSIZE * abs(inc_x); i++) {
+            x[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+        }
+
+        for (l = 0; l < loops; l++) {
+            clock_gettime(CLOCK_REALTIME, &start);
+            TRMV (&uplo, &trans, &diag, &n, a, &n, x, &inc_x);
+            clock_gettime(CLOCK_REALTIME, &stop);
+
+            time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_nsec - start.tv_nsec)) / 1.e9;
+            timeg += time1;
+        }
+
+        timeg /= loops;
+        fprintf(stderr, " %10.2f MFlops %12.9f sec\n",
+                COMPSIZE * COMPSIZE * 1. * (double)n * (double)n / timeg / 1.e6, timeg);
+    }
+
+    return 0;
+}
+
+// void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));
diff --git a/benchmark/trsm.c b/benchmark/trsm.c
index 9eae3380c..6ce1d532c 100644
--- a/benchmark/trsm.c
+++ b/benchmark/trsm.c
@@ -191,8 +191,8 @@ int main(int argc, char *argv[]){
 
    		 for(j = 0; j < m; j++){
       			for(i = 0; i < m * COMPSIZE; i++){
-				a[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
-				b[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+				a[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+				b[(long)i + (long)j * (long)m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
       		 	}
     		 }
 
diff --git a/benchmark/trsv.c b/benchmark/trsv.c
new file mode 100644
index 000000000..c60890de4
--- /dev/null
+++ b/benchmark/trsv.c
@@ -0,0 +1,222 @@
+/***************************************************************************
+Copyright (c) 2014, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include <stdio.h>
+#include <stdlib.h>
+#ifdef __CYGWIN32__
+#include <sys/time.h>
+#endif
+#include <time.h>
+#include "common.h"
+
+
+#undef GEMV
+#undef TRSV
+
+#ifndef COMPLEX
+
+#ifdef DOUBLE
+#define TRSV   BLASFUNC(dtrsv)
+#else
+#define TRSV   BLASFUNC(strsv)
+#endif
+
+#else
+
+#ifdef DOUBLE
+#define TRSV   BLASFUNC(ztrsv)
+#else
+#define TRSV   BLASFUNC(ctrsv)
+#endif
+
+#endif
+
+#if defined(__WIN32__) || defined(__WIN64__)
+
+#ifndef DELTA_EPOCH_IN_MICROSECS
+#define DELTA_EPOCH_IN_MICROSECS 11644473600000000ULL
+#endif
+
+int gettimeofday(struct timeval *tv, void *tz){
+
+  FILETIME ft;
+  unsigned __int64 tmpres = 0;
+  static int tzflag;
+
+  if (NULL != tv)
+    {
+      GetSystemTimeAsFileTime(&ft);
+
+      tmpres |= ft.dwHighDateTime;
+      tmpres <<= 32;
+      tmpres |= ft.dwLowDateTime;
+
+      /*converting file time to unix epoch*/
+      tmpres /= 10;  /*convert into microseconds*/
+      tmpres -= DELTA_EPOCH_IN_MICROSECS;
+      tv->tv_sec = (long)(tmpres / 1000000UL);
+      tv->tv_usec = (long)(tmpres % 1000000UL);
+    }
+
+  return 0;
+}
+
+#endif
+
+#if !defined(__WIN32__) && !defined(__WIN64__) && !defined(__CYGWIN32__) && 0
+
+static void *huge_malloc(BLASLONG size){
+  int shmid;
+  void *address;
+
+#ifndef SHM_HUGETLB
+#define SHM_HUGETLB 04000
+#endif
+
+  if ((shmid =shmget(IPC_PRIVATE,
+		     (size + HUGE_PAGESIZE) & ~(HUGE_PAGESIZE - 1),
+		     SHM_HUGETLB | IPC_CREAT |0600)) < 0) {
+    printf( "Memory allocation failed(shmget).\n");
+    exit(1);
+  }
+
+  address = shmat(shmid, NULL, SHM_RND);
+
+  if ((BLASLONG)address == -1){
+    printf( "Memory allocation failed(shmat).\n");
+    exit(1);
+  }
+
+  shmctl(shmid, IPC_RMID, 0);
+
+  return address;
+}
+
+#define malloc huge_malloc
+
+#endif
+
+int main(int argc, char *argv[]){
+
+  FLOAT *a, *x;
+  blasint n = 0, i, j;
+  blasint inc_x=1;
+  int loops = 1;
+  int l;
+  char *p;
+
+  int from =   1;
+  int to   = 200;
+  int step =   1;
+
+  struct timespec time_start, time_end;
+  time_t seconds = 0;
+
+  double time1,timeg;
+  long long nanos = 0;
+
+  argc--;argv++;
+
+  if (argc > 0) { from     = atol(*argv);		argc--; argv++;}
+  if (argc > 0) { to       = MAX(atol(*argv), from);	argc--; argv++;}
+  if (argc > 0) { step     = atol(*argv);		argc--; argv++;}
+
+  char uplo ='L';
+  char transa = 'N';
+  char diag ='U';
+
+  if ((p = getenv("OPENBLAS_LOOPS")))  loops = atoi(p);
+  if ((p = getenv("OPENBLAS_INCX")))   inc_x = atoi(p);
+  if ((p = getenv("OPENBLAS_TRANSA")))  transa=*p;
+  if ((p = getenv("OPENBLAS_DIAG")))  diag=*p;
+  if ((p = getenv("OPENBLAS_UPLO")))  uplo=*p;
+
+  fprintf(stderr, "From : %3d  To : %3d Step = %3d Transa = '%c' Inc_x = %d uplo=%c diag=%c loop = %d\n", from, to, step,transa,inc_x,
+          uplo,diag,loops);
+
+
+#ifdef linux
+  srandom(getpid());
+#endif
+
+  fprintf(stderr, "   SIZE       Flops\n");
+  fprintf(stderr, "============================================\n");
+
+  for(n = from; n <= to; n += step)
+  {
+      timeg=0;
+      if (( a = (FLOAT *)malloc(sizeof(FLOAT) * n * n * COMPSIZE)) == NULL){
+          fprintf(stderr,"Out of Memory!!\n");exit(1);
+      }
+
+      if (( x = (FLOAT *)malloc(sizeof(FLOAT) * n * abs(inc_x) * COMPSIZE)) == NULL){
+          fprintf(stderr,"Out of Memory!!\n");exit(1);
+      }
+
+      for(j = 0; j < n; j++){
+          for(i = 0; i < n * COMPSIZE; i++){
+              a[i + j * n * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+          }
+      }
+
+      for(i = 0; i < n * COMPSIZE * abs(inc_x); i++){
+          x[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
+      }
+
+      for(l =0;l< loops;l++){
+
+          clock_gettime(CLOCK_PROCESS_CPUTIME_ID,&time_start);
+
+          TRSV(&uplo,&transa,&diag,&n,a,&n,x,&inc_x);
+
+          clock_gettime(CLOCK_PROCESS_CPUTIME_ID,&time_end);
+          nanos = time_end.tv_nsec - time_start.tv_nsec;
+          seconds = time_end.tv_sec - time_start.tv_sec;
+
+          time1 = seconds + nanos /1.e9;
+          timeg += time1;
+      }
+
+
+      timeg /= loops;
+      long long muls = n*(n+1)/2.0;
+      long long adds = (n - 1.0)*n/2.0;
+
+      fprintf(stderr, "%10d   %10.2f MFlops %10.6f sec\n", n,(muls+adds) / timeg * 1.e-6, timeg);
+      if(a != NULL){
+        free(a);
+      }
+
+      if( x != NULL){
+        free(x);
+      }
+
+  }
+
+  return 0;
+}
+
diff --git a/benchmark/zdot.c b/benchmark/zdot.c
index ed9d4d2e8..136135c9c 100644
--- a/benchmark/zdot.c
+++ b/benchmark/zdot.c
@@ -170,9 +170,11 @@ int main(int argc, char *argv[]){
 			y[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
    	}
     	gettimeofday( &start, (struct timezone *)0);
-
+#ifdef RETURN_BY_STACK
+    	DOT (&result , &m, x, &inc_x, y, &inc_y );
+#else
     	result = DOT (&m, x, &inc_x, y, &inc_y );
-
+#endif
     	gettimeofday( &stop, (struct timezone *)0);
 
     	time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
diff --git a/c_check b/c_check
index c7899c84f..8234c2081 100644
--- a/c_check
+++ b/c_check
@@ -310,6 +310,7 @@ $linker_a = "";
 	    && ($flags !~ /advapi32/)
 	    && ($flags !~ /shell32/)
 	    && ($flags !~ /omp/)
+	    && ($flags !~ /[0-9]+/)
 	    ) {
 	    $linker_l .= $flags . " "
 	}
diff --git a/cblas.h b/cblas.h
index 1a87074d6..4bc5588d8 100644
--- a/cblas.h
+++ b/cblas.h
@@ -25,6 +25,11 @@ char* openblas_get_config(void);
 /*Get the CPU corename on runtime.*/
 char* openblas_get_corename(void);
 
+#ifdef OPENBLAS_OS_LINUX
+/* Sets thread affinity for OpenBLAS threads. `thread_idx` is in [0, openblas_get_num_threads()-1]. */
+int openblas_setaffinity(int thread_idx, size_t cpusetsize, cpu_set_t* cpu_set);
+#endif
+
 /* Get the parallelization type which is used by OpenBLAS */
 int openblas_get_parallel(void);
 /* OpenBLAS is compiled for sequential use  */
diff --git a/cmake/cc.cmake b/cmake/cc.cmake
index 22217575c..d5551147c 100644
--- a/cmake/cc.cmake
+++ b/cmake/cc.cmake
@@ -99,7 +99,7 @@ endif ()
 if (${CORE} STREQUAL "SKYLAKEX")
   if (NOT DYNAMIC_ARCH)
     if (NOT NO_AVX512)
-      set (CCOMMON_OPT = "${CCOMMON_OPT} -march=skylake-avx512")
+      set (CCOMMON_OPT "${CCOMMON_OPT} -march=skylake-avx512")
     endif ()
   endif ()
 endif ()
diff --git a/cmake/fc.cmake b/cmake/fc.cmake
index f54c989d4..fc1f9bb22 100644
--- a/cmake/fc.cmake
+++ b/cmake/fc.cmake
@@ -16,6 +16,7 @@ if (${F_COMPILER} STREQUAL "FLANG")
   if (USE_OPENMP)
     set(FCOMMON_OPT "${FCOMMON_OPT} -fopenmp")
   endif ()
+  set(FCOMMON_OPT "${FCOMMON_OPT} -Mrecursive -Kieee")
 endif ()
 
 if (${F_COMPILER} STREQUAL "G77")
@@ -81,6 +82,7 @@ if (${F_COMPILER} STREQUAL "INTEL")
   if (INTERFACE64)
     set(FCOMMON_OPT "${FCOMMON_OPT} -i8")
   endif ()
+  set(FCOMMON_OPT "${FCOMMON_OPT} -recursive")
   if (USE_OPENMP)
     set(FCOMMON_OPT "${FCOMMON_OPT} -openmp")
   endif ()
@@ -120,6 +122,7 @@ if (${F_COMPILER} STREQUAL "PGI")
   else ()
     set(FCOMMON_OPT "${FCOMMON_OPT} -tp p7")
   endif ()
+  set(FCOMMON_OPT "${FCOMMON_OPT} -Mrecursive")
   if (USE_OPENMP)
     set(FCOMMON_OPT "${FCOMMON_OPT} -mp")
   endif ()
diff --git a/cmake/kernel.cmake b/cmake/kernel.cmake
index 9b238f004..4b505a102 100644
--- a/cmake/kernel.cmake
+++ b/cmake/kernel.cmake
@@ -113,11 +113,31 @@ macro(SetDefaultL1)
   set(ZSUMKERNEL zsum.S)
   set(QSUMKERNEL sum.S)
   set(XSUMKERNEL zsum.S)
+if (BUILD_HALF)
+  set(SHAMINKERNEL ../arm/amin.c)
+  set(SHAMAXKERNEL ../arm/amax.c)
+  set(SHMAXKERNEL ../arm/max.c)
+  set(SHMINKERNEL ../arm/min.c)
+  set(ISHAMAXKERNEL ../arm/iamax.c)
+  set(ISHAMINKERNEL ../arm/iamin.c)
+  set(ISHMAXKERNEL ../arm/imax.c)
+  set(ISHMINKERNEL ../arm/imin.c)
+  set(SHASUMKERNEL ../arm/asum.c)
+  set(SHAXPYKERNEL ../arm/axpy.c)
+  set(SHAXPBYKERNEL ../arm/axpby.c)
+  set(SHCOPYKERNEL ../arm/copy.c)
+  set(SHDOTKERNEL ../arm/dot.c)
+  set(SHROTKERNEL ../arm/rot.c)
+  set(SHSCALKERNEL ../arm/scal.c)
+  set(SHNRM2KERNEL ../arm/nrm2.c)
+  set(SHSUMKERNEL ../arm/sum.c)
+  set(SHSWAPKERNEL ../arm/swap.c)
+endif ()
 endmacro ()
 
 macro(SetDefaultL2)
-  set(SGEMVNKERNEL gemv_n.S)
-  set(SGEMVTKERNEL gemv_t.S)
+  set(SGEMVNKERNEL ../arm/gemv_n.c)
+  set(SGEMVTKERNEL ../arm/gemv_t.c)
   set(DGEMVNKERNEL gemv_n.S)
   set(DGEMVTKERNEL gemv_t.S)
   set(CGEMVNKERNEL zgemv_n.S)
@@ -161,6 +181,11 @@ macro(SetDefaultL2)
   set(XHEMV_L_KERNEL ../generic/zhemv_k.c)
   set(XHEMV_V_KERNEL ../generic/zhemv_k.c)
   set(XHEMV_M_KERNEL ../generic/zhemv_k.c)
+if (BUILD_HALF)
+  set(SHGEMVNKERNEL ../arm/gemv_n.c)
+  set(SHGEMVTKERNEL ../arm/gemv_t.c)
+  set(SHGERKERNEL ../generic/ger.c)
+endif ()
 endmacro ()
 
 macro(SetDefaultL3)
@@ -168,4 +193,18 @@ macro(SetDefaultL3)
   set(DGEADD_KERNEL ../generic/geadd.c)
   set(CGEADD_KERNEL ../generic/zgeadd.c)
   set(ZGEADD_KERNEL ../generic/zgeadd.c)
+if (BUILD_HALF)
+  set(SHGEADD_KERNEL ../generic/geadd.c)
+  set(SHGEMMKERNEL ../generic/gemmkernel_2x2.c)
+  set(SHGEMM_BETA  ../generic/gemm_beta.c)
+  set(SHGEMMINCOPY ../generic/gemm_ncopy_2.c)
+  set(SHGEMMITCOPY ../generic/gemm_tcopy_2.c)
+  set(SHGEMMONCOPY ../generic/gemm_ncopy_2.c)
+  set(SHGEMMOTCOPY ../generic/gemm_tcopy_2.c)
+  set(SHGEMMINCOPYOBJ shgemm_incopy.o)
+  set(SHGEMMITCOPYOBJ shgemm_itcopy.o)
+  set(SHGEMMONCOPYOBJ shgemm_oncopy.o)
+  set(SHGEMMOTCOPYOBJ shgemm_otcopy.o)
+endif ()
+
 endmacro ()
diff --git a/cmake/os.cmake b/cmake/os.cmake
index 2d25e7aaa..c644bc3f7 100644
--- a/cmake/os.cmake
+++ b/cmake/os.cmake
@@ -8,7 +8,7 @@ if (${CMAKE_SYSTEM_NAME} STREQUAL "Linux")
   set(NO_EXPRECISION 1)
 endif ()
 
-if (${CMAKE_SYSTEM_NAME} MATCHES "FreeBSD|OpenBSD|NetBSD|DragonFly")
+if (${CMAKE_SYSTEM_NAME} MATCHES "FreeBSD|OpenBSD|NetBSD|DragonFly|Darwin")
   set(EXTRALIB "${EXTRALIB} -lm")
   set(NO_EXPRECISION 1)
 endif ()
diff --git a/cmake/prebuild.cmake b/cmake/prebuild.cmake
index 44e1473d1..067b97b4b 100644
--- a/cmake/prebuild.cmake
+++ b/cmake/prebuild.cmake
@@ -16,6 +16,8 @@
 # HAVE_SSE2
 # HAVE_SSE3
 # MAKE
+# SHGEMM_UNROLL_M
+# SHGEMM_UNROLL_N
 # SGEMM_UNROLL_M
 # SGEMM_UNROLL_N
 # DGEMM_UNROLL_M
@@ -437,6 +439,8 @@ if (DEFINED CORE AND CMAKE_CROSSCOMPILING AND NOT (${HOST_OS} STREQUAL "WINDOWSS
     set(ZGEMM_UNROLL_N 2)
     set(SYMV_P 8)
   endif()
+  set(SHGEMM_UNROLL_M 8)
+  set(SHGEMM_UNROLL_N 4)
 
   # Or should this actually be NUM_CORES?
   if (${NUM_THREADS} GREATER 0)
@@ -488,7 +492,7 @@ else(NOT CMAKE_CROSSCOMPILING)
   if (NOT "${CMAKE_SYSTEM_NAME}" STREQUAL "WindowsStore")
     try_compile(GETARCH_RESULT ${GETARCH_DIR}
       SOURCES ${GETARCH_SRC}
-    COMPILE_DEFINITIONS ${EXFLAGS} ${GETARCH_FLAGS} -I${GETARCH_DIR} -I"${PROJECT_SOURCE_DIR}" -I"${PROJECT_BINARY_DIR}"
+    COMPILE_DEFINITIONS ${EXFLAGS} ${GETARCH_FLAGS} -I"${GETARCH_DIR}" -I"${PROJECT_SOURCE_DIR}" -I"${PROJECT_BINARY_DIR}"
       OUTPUT_VARIABLE GETARCH_LOG
       COPY_FILE ${PROJECT_BINARY_DIR}/${GETARCH_BIN}
     )
@@ -516,7 +520,7 @@ execute_process(COMMAND "${PROJECT_BINARY_DIR}/${GETARCH_BIN}" 1 OUTPUT_VARIABLE
   if (NOT "${CMAKE_SYSTEM_NAME}" STREQUAL "WindowsStore")
     try_compile(GETARCH2_RESULT ${GETARCH2_DIR}
       SOURCES ${PROJECT_SOURCE_DIR}/getarch_2nd.c
-    COMPILE_DEFINITIONS ${EXFLAGS} ${GETARCH_FLAGS} ${GETARCH2_FLAGS} -I${GETARCH2_DIR} -I"${PROJECT_SOURCE_DIR}" -I"${PROJECT_BINARY_DIR}"
+    COMPILE_DEFINITIONS ${EXFLAGS} ${GETARCH_FLAGS} ${GETARCH2_FLAGS} -I"${GETARCH2_DIR}" -I"${PROJECT_SOURCE_DIR}" -I"${PROJECT_BINARY_DIR}"
       OUTPUT_VARIABLE GETARCH2_LOG
       COPY_FILE ${PROJECT_BINARY_DIR}/${GETARCH2_BIN}
     )
diff --git a/cmake/system.cmake b/cmake/system.cmake
index ce980a7b9..d8dcc3cf3 100644
--- a/cmake/system.cmake
+++ b/cmake/system.cmake
@@ -297,6 +297,16 @@ if (USE_SIMPLE_THREADED_LEVEL3)
   set(CCOMMON_OPT "${CCOMMON_OPT} -DUSE_SIMPLE_THREADED_LEVEL3")
 endif ()
 
+if (NOT ${CMAKE_SYSTEM_NAME} STREQUAL "Windows")
+if (DEFINED MAX_STACK_ALLOC)
+if (NOT ${MAX_STACK_ALLOC} EQUAL 0)
+set(CCOMMON_OPT "${CCOMMON_OPT} -DMAX_STACK_ALLOC=${MAX_STACK_ALLOC}")
+endif ()
+else ()
+set(CCOMMON_OPT "${CCOMMON_OPT} -DMAX_STACK_ALLOC=2048")
+endif ()
+endif ()
+
 if (DEFINED LIBNAMESUFFIX)
   set(LIBPREFIX "libopenblas_${LIBNAMESUFFIX}")
 else ()
@@ -407,6 +417,14 @@ if (${CMAKE_C_COMPILER} STREQUAL "LSB" OR ${CMAKE_SYSTEM_NAME} STREQUAL "Windows
   set(LAPACK_CFLAGS "${LAPACK_CFLAGS} -DLAPACK_COMPLEX_STRUCTURE")
 endif ()
 
+if ("${CMAKE_BUILD_TYPE}" STREQUAL "Release")
+if ("${F_COMPILER}" STREQUAL "FLANG")
+if (${CMAKE_Fortran_COMPILER_VERSION} VERSION_LESS_EQUAL 3)
+  set(CMAKE_Fortran_FLAGS_RELEASE "${CMAKE_Fortran_FLAGS_RELEASE} -fno-unroll-loops")
+endif ()
+endif ()
+endif ()
+
 if (NOT DEFINED SUFFIX)
   set(SUFFIX o)
 endif ()
@@ -530,6 +548,8 @@ endif ()
 #export FUNCTION_PROFILE
 #export TARGET_CORE
 #
+#export SHGEMM_UNROLL_M
+#export SHGEMM_UNROLL_N
 #export SGEMM_UNROLL_M
 #export SGEMM_UNROLL_N
 #export DGEMM_UNROLL_M
diff --git a/cmake/system_check.cmake b/cmake/system_check.cmake
index c4a553c5a..94eb0a9c6 100644
--- a/cmake/system_check.cmake
+++ b/cmake/system_check.cmake
@@ -109,10 +109,10 @@ else()
 endif()
 
 if (X86_64 OR X86)
-  file(WRITE ${PROJECT_BINARY_DIR}/avx512.tmp "#include <immintrin.h>\n\nint main(void){ __asm__ volatile(\"vbroadcastss -4 * 4(%rsi), %zmm2\"); }")
-execute_process(COMMAND ${CMAKE_C_COMPILER} -march=skylake-avx512 -c -v -o ${PROJECT_BINARY_DIR}/avx512.o -x c ${PROJECT_BINARY_DIR}/avx512.tmp OUTPUT_QUIET ERROR_QUIET RESULT_VARIABLE NO_AVX512)
+  file(WRITE ${PROJECT_BINARY_DIR}/avx512.c "#include <immintrin.h>\n\nint main(void){ __asm__ volatile(\"vbroadcastss -4 * 4(%rsi), %zmm2\"); }")
+execute_process(COMMAND ${CMAKE_C_COMPILER} -march=skylake-avx512 -c -v -o ${PROJECT_BINARY_DIR}/avx512.o ${PROJECT_BINARY_DIR}/avx512.c OUTPUT_QUIET ERROR_QUIET RESULT_VARIABLE NO_AVX512)
 if (NO_AVX512 EQUAL 1)
 set (CCOMMON_OPT "${CCOMMON_OPT} -DNO_AVX512")
 endif()
-  file(REMOVE "avx512.tmp" "avx512.o")
+  file(REMOVE "avx512.c" "avx512.o")
 endif()
diff --git a/cmake/utils.cmake b/cmake/utils.cmake
index fd93f8a70..1c21e776e 100644
--- a/cmake/utils.cmake
+++ b/cmake/utils.cmake
@@ -15,12 +15,36 @@ endfunction ()
 # Reads a Makefile into CMake vars.
 macro(ParseMakefileVars MAKEFILE_IN)
   message(STATUS "Reading vars from ${MAKEFILE_IN}...")
+        set (IfElse 0)
+        set (ElseSeen 0)
   file(STRINGS ${MAKEFILE_IN} makefile_contents)
   foreach (makefile_line ${makefile_contents})
+#message(STATUS "parsing ${makefile_line}")
+    if (${IfElse} GREATER 0)
+      string(REGEX MATCH "endif[ \t]*" line_match "${makefile_line}")
+      if (NOT "${line_match}" STREQUAL "")
+#           message(STATUS "ENDIF ${makefile_line}")
+        set (IfElse 0)
+        set (ElseSeen 0)
+        continue ()
+      endif ()
+      string(REGEX MATCH "else[ \t]*" line_match "${makefile_line}")
+        if (NOT "${line_match}" STREQUAL "") 
+#           message(STATUS "ELSE ${makefile_line}")
+           set (ElseSeen 1)        
+           continue ()  
+        endif()
+      if ( (${IfElse} EQUAL 2 AND ${ElseSeen} EQUAL 0) OR ( ${IfElse} EQUAL 1 AND ${ElseSeen} EQUAL 1))
+#           message(STATUS "skipping ${makefile_line}")
+         continue ()
+      endif ()    
+    endif ()    
     string(REGEX MATCH "([0-9_a-zA-Z]+)[ \t]*=[ \t]*(.+)$" line_match "${makefile_line}")
     if (NOT "${line_match}" STREQUAL "")
+#message(STATUS "match on ${line_match}")
       set(var_name ${CMAKE_MATCH_1})
-      set(var_value ${CMAKE_MATCH_2})
+#      set(var_value ${CMAKE_MATCH_2})
+      string(STRIP ${CMAKE_MATCH_2} var_value)
       # check for Makefile variables in the string, e.g. $(TSUFFIX)
       string(REGEX MATCHALL "\\$\\(([0-9_a-zA-Z]+)\\)" make_var_matches ${var_value})
       foreach (make_var ${make_var_matches})
@@ -33,7 +57,31 @@ macro(ParseMakefileVars MAKEFILE_IN)
     else ()
       string(REGEX MATCH "include \\$\\(KERNELDIR\\)/(.+)$" line_match "${makefile_line}")
       if (NOT "${line_match}" STREQUAL "")
+#message(STATUS "match on include ${line_match}")
         ParseMakefileVars(${KERNELDIR}/${CMAKE_MATCH_1})
+      else ()
+#        message(STATUS "unmatched line ${line_match}")
+        string(REGEX MATCH "ifeq \\(\\$\\(([_A-Z]+)\\),[ \t]*([0-9_A-Z]+)\\)" line_match "${makefile_line}")
+        if (NOT "${line_match}" STREQUAL "")
+#          message(STATUS "IFEQ: ${line_match} first: ${CMAKE_MATCH_1} second: ${CMAKE_MATCH_2}")
+          if (DEFINED ${${CMAKE_MATCH_1}} AND ${${CMAKE_MATCH_1}} STREQUAL ${CMAKE_MATCH_2})
+#            message (STATUS "condition is true")
+            set (IfElse 1)
+          else ()
+            set (IfElse 2)
+          endif ()
+        else ()
+          string(REGEX MATCH "ifneq \\(\\$\\(([_A-Z]+)\\),[ \t]*([0-9_A-Z]+)\\)" line_match "${makefile_line}")
+          if (NOT "${line_match}" STREQUAL "")
+#            message(STATUS "IFNEQ: ${line_match} first: ${CMAKE_MATCH_1} second: ${CMAKE_MATCH_2}")
+            if (NOT ( ${${CMAKE_MATCH_1}} STREQUAL ${CMAKE_MATCH_2}))
+#              message (STATUS "condition is true")
+              set (IfElse 1)
+            else ()
+              set (IfElse 2)
+            endif ()
+          endif ()
+        endif ()
       endif ()
     endif ()
   endforeach ()
@@ -163,6 +211,7 @@ function(GenerateNamedObjects sources_in)
   if (complex_only)
     list(REMOVE_ITEM float_list "SINGLE")
     list(REMOVE_ITEM float_list "DOUBLE")
+    list(REMOVE_ITEM float_list "HALF")
   elseif (real_only)
     list(REMOVE_ITEM float_list "COMPLEX")
     list(REMOVE_ITEM float_list "ZCOMPLEX")
@@ -176,6 +225,9 @@ function(GenerateNamedObjects sources_in)
       if (NOT no_float_type)
         string(SUBSTRING ${float_type} 0 1 float_char)
         string(TOLOWER ${float_char} float_char)
+	if (${float_type} STREQUAL "HALF")
+		set (float_char "sh")
+	endif ()
       endif ()
 
       if (NOT name_in)
@@ -210,6 +262,9 @@ function(GenerateNamedObjects sources_in)
       if (${float_type} STREQUAL "DOUBLE" OR ${float_type} STREQUAL "ZCOMPLEX")
         list(APPEND obj_defines "DOUBLE")
       endif ()
+      if (${float_type} STREQUAL "HALF")
+        list(APPEND obj_defines "HALF")
+      endif ()
       if (${float_type} STREQUAL "COMPLEX" OR ${float_type} STREQUAL "ZCOMPLEX")
         list(APPEND obj_defines "COMPLEX")
         if (mangle_complex_sources)
diff --git a/common.h b/common.h
index a9fe8d911..e2c8cdee5 100644
--- a/common.h
+++ b/common.h
@@ -257,6 +257,11 @@ typedef long BLASLONG;
 typedef unsigned long BLASULONG;
 #endif
 
+#ifndef BFLOAT16
+typedef unsigned short bfloat16;
+#define HALFCONVERSION 1
+#endif
+
 #ifdef USE64BITINT
 typedef BLASLONG blasint;
 #if defined(OS_WINDOWS) && defined(__64BIT__)
@@ -297,6 +302,13 @@ typedef int blasint;
 #define SIZE	8
 #define  BASE_SHIFT 3
 #define ZBASE_SHIFT 4
+#elif defined(HALF)
+#define IFLOAT	bfloat16
+#define XFLOAT IFLOAT
+#define FLOAT	float
+#define SIZE   2
+#define BASE_SHIFT 1
+#define ZBASE_SHIFT 2
 #else
 #define FLOAT	float
 #define SIZE    4
@@ -308,6 +320,10 @@ typedef int blasint;
 #define XFLOAT	FLOAT
 #endif
 
+#ifndef IFLOAT
+#define IFLOAT	FLOAT
+#endif
+
 #ifndef COMPLEX
 #define COMPSIZE  1
 #else
@@ -657,6 +673,8 @@ void gotoblas_dynamic_init(void);
 void gotoblas_dynamic_quit(void);
 void gotoblas_profile_init(void);
 void gotoblas_profile_quit(void);
+	
+int support_avx512(void);	
 
 #ifdef USE_OPENMP
 
diff --git a/common_alpha.h b/common_alpha.h
index 9739c941d..f1ea8ff94 100644
--- a/common_alpha.h
+++ b/common_alpha.h
@@ -43,6 +43,7 @@
 
 #define MB  asm("mb")
 #define WMB asm("wmb")
+#define RMB asm("rmb")
 
 static void __inline blas_lock(unsigned long *address){
 #ifndef __DECC
diff --git a/common_arm.h b/common_arm.h
index 27fa76b76..682315de5 100644
--- a/common_arm.h
+++ b/common_arm.h
@@ -37,11 +37,13 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define MB
 #define WMB
+#define RMB
 
 #else
 
 #define MB   __asm__ __volatile__ ("dmb  ish" : : : "memory")
 #define WMB  __asm__ __volatile__ ("dmb  ishst" : : : "memory")
+#define RMB  __asm__ __volatile__ ("dmb  ish" : : : "memory")
 
 #endif
 
@@ -121,7 +123,7 @@ REALNAME:
 #endif
 #define HUGE_PAGESIZE   ( 4 << 20)
 
-#define BUFFER_SIZE     (16 << 20)
+#define BUFFER_SIZE     (32 << 20)
 
 
 #define BASE_ADDRESS (START_ADDRESS - BUFFER_SIZE * MAX_CPU_NUMBER)
diff --git a/common_arm64.h b/common_arm64.h
index a928dbe7b..314946282 100644
--- a/common_arm64.h
+++ b/common_arm64.h
@@ -35,7 +35,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define MB   __asm__ __volatile__ ("dmb  ish" : : : "memory")
 #define WMB  __asm__ __volatile__ ("dmb  ishst" : : : "memory")
-
+#define RMB  __asm__ __volatile__ ("dmb  ishld" : : : "memory")
 
 #define INLINE inline
 
@@ -141,12 +141,17 @@ REALNAME:
 #endif
 #define HUGE_PAGESIZE   ( 4 << 20)
 
+#ifndef BUFFERSIZE
 #if defined(CORTEXA57)
 #define BUFFER_SIZE     (20 << 20)
+#elif defined(TSV110) || defined(EMAG8180)
+#define BUFFER_SIZE     (32 << 20)
 #else
 #define BUFFER_SIZE     (16 << 20)
 #endif
-
+#else
+#define BUFFER_SIZE	(32 << BUFFERSIZE)
+#endif
 
 #define BASE_ADDRESS (START_ADDRESS - BUFFER_SIZE * MAX_CPU_NUMBER)
 
diff --git a/common_ia64.h b/common_ia64.h
index 72b75fc4e..59aefbd6d 100644
--- a/common_ia64.h
+++ b/common_ia64.h
@@ -47,6 +47,7 @@
 
 #define MB
 #define WMB
+#define RMB
 
 #ifdef __ECC
 #include <ia64intrin.h>
diff --git a/common_interface.h b/common_interface.h
index c350ac8ec..78f5be6b0 100644
--- a/common_interface.h
+++ b/common_interface.h
@@ -469,6 +469,8 @@ void BLASFUNC(xhbmv)(char *, blasint *, blasint *, xdouble *, xdouble *, blasint
 
 /* Level 3 routines */
 
+void BLASFUNC(shgemm)(char *, char *, blasint *, blasint *, blasint *, float *,
+	   bfloat16 *, blasint *, bfloat16 *, blasint *, float *, float *, blasint *);
 void BLASFUNC(sgemm)(char *, char *, blasint *, blasint *, blasint *, float *,
 	   float  *, blasint *, float  *, blasint *, float  *, float  *, blasint *);
 void BLASFUNC(dgemm)(char *, char *, blasint *, blasint *, blasint *, double *,
diff --git a/common_level3.h b/common_level3.h
index 6fa902be8..4e44a5e73 100644
--- a/common_level3.h
+++ b/common_level3.h
@@ -55,6 +55,8 @@ extern void sgemm_kernel_direct(BLASLONG M, BLASLONG N, BLASLONG K,
 extern int sgemm_kernel_direct_performant(BLASLONG M, BLASLONG N, BLASLONG K);
 
 
+int shgemm_beta(BLASLONG, BLASLONG, BLASLONG, float,
+	       bfloat16 *, BLASLONG, bfloat16 *, BLASLONG, float *, BLASLONG);
 int sgemm_beta(BLASLONG, BLASLONG, BLASLONG, float,
 	       float  *, BLASLONG, float   *, BLASLONG, float  *, BLASLONG);
 int dgemm_beta(BLASLONG, BLASLONG, BLASLONG, double,
@@ -76,6 +78,10 @@ int xgemm_beta(BLASLONG, BLASLONG, BLASLONG, xdouble *,
 	       xdouble *, BLASLONG, xdouble  *, BLASLONG, xdouble *, BLASLONG);
 #endif
 
+int shgemm_incopy(BLASLONG m, BLASLONG n, bfloat16 *a, BLASLONG lda, bfloat16 *b);
+int shgemm_itcopy(BLASLONG m, BLASLONG n, bfloat16 *a, BLASLONG lda, bfloat16 *b);
+int shgemm_oncopy(BLASLONG m, BLASLONG n, bfloat16 *a, BLASLONG lda, bfloat16 *b);
+int shgemm_otcopy(BLASLONG m, BLASLONG n, bfloat16 *a, BLASLONG lda, bfloat16 *b);
 int sgemm_incopy(BLASLONG m, BLASLONG n, float *a, BLASLONG lda, float *b);
 int sgemm_itcopy(BLASLONG m, BLASLONG n, float *a, BLASLONG lda, float *b);
 int sgemm_oncopy(BLASLONG m, BLASLONG n, float *a, BLASLONG lda, float *b);
@@ -499,6 +505,7 @@ int xher2k_kernel_UC(BLASLONG m, BLASLONG n, BLASLONG k, xdouble alpha_r, xdoubl
 int xher2k_kernel_LN(BLASLONG m, BLASLONG n, BLASLONG k, xdouble alpha_r, xdouble alpha_i, xdouble *a, xdouble *b, xdouble *c, BLASLONG ldc, BLASLONG offset, int flag);
 int xher2k_kernel_LC(BLASLONG m, BLASLONG n, BLASLONG k, xdouble alpha_r, xdouble alpha_i, xdouble *a, xdouble *b, xdouble *c, BLASLONG ldc, BLASLONG offset, int flag);
 
+int shgemm_kernel(BLASLONG, BLASLONG, BLASLONG, float,  bfloat16 *, bfloat16 *, float *, BLASLONG);
 int sgemm_kernel(BLASLONG, BLASLONG, BLASLONG, float,  float  *, float  *, float  *, BLASLONG);
 int dgemm_kernel(BLASLONG, BLASLONG, BLASLONG, double, double *, double *, double *, BLASLONG);
 
@@ -527,6 +534,11 @@ int cgemm3m_kernel(BLASLONG, BLASLONG, BLASLONG, float,  float,  float  *, float
 int zgemm3m_kernel(BLASLONG, BLASLONG, BLASLONG, double, double, double *, double *, double *, BLASLONG);
 int xgemm3m_kernel(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG);
 
+int shgemm_nn(blas_arg_t *, BLASLONG *, BLASLONG *, bfloat16 *, bfloat16 *, BLASLONG);
+int shgemm_nt(blas_arg_t *, BLASLONG *, BLASLONG *, bfloat16 *, bfloat16 *, BLASLONG);
+int shgemm_tn(blas_arg_t *, BLASLONG *, BLASLONG *, bfloat16 *, bfloat16 *, BLASLONG);
+int shgemm_tt(blas_arg_t *, BLASLONG *, BLASLONG *, bfloat16 *, bfloat16 *, BLASLONG);
+
 int sgemm_nn(blas_arg_t *, BLASLONG *, BLASLONG *, float *, float *, BLASLONG);
 int sgemm_nt(blas_arg_t *, BLASLONG *, BLASLONG *, float *, float *, BLASLONG);
 int sgemm_tn(blas_arg_t *, BLASLONG *, BLASLONG *, float *, float *, BLASLONG);
@@ -619,6 +631,11 @@ int xgemm_cr(blas_arg_t *, BLASLONG *, BLASLONG *, xdouble *, xdouble *, BLASLON
 int xgemm_cc(blas_arg_t *, BLASLONG *, BLASLONG *, xdouble *, xdouble *, BLASLONG);
 #endif
 
+int shgemm_thread_nn(blas_arg_t *, BLASLONG *, BLASLONG *, bfloat16 *, bfloat16 *, BLASLONG);
+int shgemm_thread_nt(blas_arg_t *, BLASLONG *, BLASLONG *, bfloat16 *, bfloat16 *, BLASLONG);
+int shgemm_thread_tn(blas_arg_t *, BLASLONG *, BLASLONG *, bfloat16 *, bfloat16 *, BLASLONG);
+int shgemm_thread_tt(blas_arg_t *, BLASLONG *, BLASLONG *, bfloat16 *, bfloat16 *, BLASLONG);
+
 int sgemm_thread_nn(blas_arg_t *, BLASLONG *, BLASLONG *, float *, float *, BLASLONG);
 int sgemm_thread_nt(blas_arg_t *, BLASLONG *, BLASLONG *, float *, float *, BLASLONG);
 int sgemm_thread_tn(blas_arg_t *, BLASLONG *, BLASLONG *, float *, float *, BLASLONG);
diff --git a/common_macro.h b/common_macro.h
index 13bb85794..8fe1f156f 100644
--- a/common_macro.h
+++ b/common_macro.h
@@ -39,6 +39,7 @@
 #ifndef COMMON_MACRO
 #define COMMON_MACRO
 
+#include "common_sh.h"
 #include "common_s.h"
 #include "common_d.h"
 #include "common_q.h"
@@ -642,6 +643,288 @@
 #define IMATCOPY_K_RT		DIMATCOPY_K_RT
 
 #define GEADD_K                 DGEADD_K
+
+#elif defined(HALF)
+
+#define	AMAX_K			SAMAX_K
+#define	AMIN_K			SAMIN_K
+#define	MAX_K			SMAX_K
+#define	MIN_K			SMIN_K
+#define	IAMAX_K			ISAMAX_K
+#define	IAMIN_K			ISAMIN_K
+#define	IMAX_K			ISMAX_K
+#define	IMIN_K			ISMIN_K
+#define	ASUM_K			SASUM_K
+#define	DOTU_K			SDOTU_K
+#define	DOTC_K			SDOTC_K
+#define	AXPYU_K			SAXPYU_K
+#define	AXPYC_K			SAXPYC_K
+#define AXPBY_K     SAXPBY_K
+#define SCAL_K			SSCAL_K
+#define GEMV_N			SGEMV_N
+#define GEMV_T			SGEMV_T
+#define SYMV_U			SSYMV_U
+#define SYMV_L			SSYMV_L
+#define	GERU_K			SGERU_K
+#define	GERC_K			SGERC_K
+#define	GERV_K			SGERV_K
+#define	GERD_K			SGERD_K
+#define	SUM_K			SSUM_K
+#define	SWAP_K			SSWAP_K
+#define	ROT_K			SROT_K
+#define COPY_K    SCOPY_K
+#define NRM2_K    SNRM2_K
+#define SYMV_THREAD_U		SSYMV_THREAD_U
+#define SYMV_THREAD_L		SSYMV_THREAD_L
+#define GEMM_BETA               SHGEMM_BETA
+#define	GEMM_KERNEL_N		SHGEMM_KERNEL
+#define	GEMM_KERNEL_L		SHGEMM_KERNEL
+#define	GEMM_KERNEL_R		SHGEMM_KERNEL
+#define	GEMM_KERNEL_B		SHGEMM_KERNEL
+
+#define	GEMM_NN			SHGEMM_NN
+#define	GEMM_CN			SHGEMM_TN
+#define	GEMM_TN			SHGEMM_TN
+#define	GEMM_NC			SHGEMM_NT
+#define	GEMM_NT			SHGEMM_NT
+#define	GEMM_CC			SHGEMM_TT
+#define	GEMM_CT			SHGEMM_TT
+#define	GEMM_TC			SHGEMM_TT
+#define	GEMM_TT			SHGEMM_TT
+#define	GEMM_NR			SHGEMM_NN
+#define	GEMM_TR			SHGEMM_TN
+#define	GEMM_CR			SHGEMM_TN
+#define	GEMM_RN			SHGEMM_NN
+#define	GEMM_RT			SHGEMM_NT
+#define	GEMM_RC			SHGEMM_NT
+#define	GEMM_RR			SHGEMM_NN
+#define	GEMM_ONCOPY		SHGEMM_ONCOPY
+#define	GEMM_OTCOPY		SHGEMM_OTCOPY
+#define	GEMM_INCOPY		SHGEMM_INCOPY
+#define	GEMM_ITCOPY		SHGEMM_ITCOPY
+#define SYMM_THREAD_LU          SSYMM_THREAD_LU
+#define SYMM_THREAD_LL          SSYMM_THREAD_LL
+#define SYMM_THREAD_RU          SSYMM_THREAD_RU
+#define SYMM_THREAD_RL          SSYMM_THREAD_RL
+#define SYMM_LU                 SSYMM_LU
+#define SYMM_LL                 SSYMM_LL
+#define SYMM_RU                 SSYMM_RU
+#define SYMM_RL                 SSYMM_RL
+
+
+#define HEMM_THREAD_LU          SHEMM_THREAD_LU
+#define HEMM_THREAD_LL          SHEMM_THREAD_LL
+#define HEMM_THREAD_RU          SHEMM_THREAD_RU
+#define HEMM_THREAD_RL          SHEMM_THREAD_RL
+
+#define	GEMM_THREAD_NN		SHGEMM_THREAD_NN
+#define	GEMM_THREAD_CN		SHGEMM_THREAD_TN
+#define	GEMM_THREAD_TN		SHGEMM_THREAD_TN
+#define	GEMM_THREAD_NC		SHGEMM_THREAD_NT
+#define	GEMM_THREAD_NT		SHGEMM_THREAD_NT
+#define	GEMM_THREAD_CC		SHGEMM_THREAD_TT
+#define	GEMM_THREAD_CT		SHGEMM_THREAD_TT
+#define	GEMM_THREAD_TC		SHGEMM_THREAD_TT
+#define	GEMM_THREAD_TT		SHGEMM_THREAD_TT
+#define	GEMM_THREAD_NR		SHGEMM_THREAD_NN
+#define	GEMM_THREAD_TR		SHGEMM_THREAD_TN
+#define	GEMM_THREAD_CR		SHGEMM_THREAD_TN
+#define	GEMM_THREAD_RN		SHGEMM_THREAD_NN
+#define	GEMM_THREAD_RT		SHGEMM_THREAD_NT
+#define	GEMM_THREAD_RC		SHGEMM_THREAD_NT
+#define	GEMM_THREAD_RR		SHGEMM_THREAD_NN
+
+#ifdef UNIT
+
+#define	TRMM_OUNCOPY		STRMM_OUNUCOPY
+#define	TRMM_OUTCOPY		STRMM_OUTUCOPY
+#define	TRMM_OLNCOPY		STRMM_OLNUCOPY
+#define	TRMM_OLTCOPY		STRMM_OLTUCOPY
+#define	TRSM_OUNCOPY		STRSM_OUNUCOPY
+#define	TRSM_OUTCOPY		STRSM_OUTUCOPY
+#define	TRSM_OLNCOPY		STRSM_OLNUCOPY
+#define	TRSM_OLTCOPY		STRSM_OLTUCOPY
+
+#define	TRMM_IUNCOPY		STRMM_IUNUCOPY
+#define	TRMM_IUTCOPY		STRMM_IUTUCOPY
+#define	TRMM_ILNCOPY		STRMM_ILNUCOPY
+#define	TRMM_ILTCOPY		STRMM_ILTUCOPY
+#define	TRSM_IUNCOPY		STRSM_IUNUCOPY
+#define	TRSM_IUTCOPY		STRSM_IUTUCOPY
+#define	TRSM_ILNCOPY		STRSM_ILNUCOPY
+#define	TRSM_ILTCOPY		STRSM_ILTUCOPY
+
+#else
+
+#define	TRMM_OUNCOPY		STRMM_OUNNCOPY
+#define	TRMM_OUTCOPY		STRMM_OUTNCOPY
+#define	TRMM_OLNCOPY		STRMM_OLNNCOPY
+#define	TRMM_OLTCOPY		STRMM_OLTNCOPY
+#define	TRSM_OUNCOPY		STRSM_OUNNCOPY
+#define	TRSM_OUTCOPY		STRSM_OUTNCOPY
+#define	TRSM_OLNCOPY		STRSM_OLNNCOPY
+#define	TRSM_OLTCOPY		STRSM_OLTNCOPY
+
+#define	TRMM_IUNCOPY		STRMM_IUNNCOPY
+#define	TRMM_IUTCOPY		STRMM_IUTNCOPY
+#define	TRMM_ILNCOPY		STRMM_ILNNCOPY
+#define	TRMM_ILTCOPY		STRMM_ILTNCOPY
+#define	TRSM_IUNCOPY		STRSM_IUNNCOPY
+#define	TRSM_IUTCOPY		STRSM_IUTNCOPY
+#define	TRSM_ILNCOPY		STRSM_ILNNCOPY
+#define	TRSM_ILTCOPY		STRSM_ILTNCOPY
+
+#define	TRMM_KERNEL_LN		STRMM_KERNEL_LN
+#define	TRMM_KERNEL_LT		STRMM_KERNEL_LT
+#define	TRMM_KERNEL_LR		STRMM_KERNEL_LN
+#define	TRMM_KERNEL_LC		STRMM_KERNEL_LT
+#define	TRMM_KERNEL_RN		STRMM_KERNEL_RN
+#define	TRMM_KERNEL_RT		STRMM_KERNEL_RT
+#define	TRMM_KERNEL_RR		STRMM_KERNEL_RN
+#define	TRMM_KERNEL_RC		STRMM_KERNEL_RT
+
+#define	TRSM_KERNEL_LN		STRSM_KERNEL_LN
+#define	TRSM_KERNEL_LT		STRSM_KERNEL_LT
+#define	TRSM_KERNEL_LR		STRSM_KERNEL_LN
+#define	TRSM_KERNEL_LC		STRSM_KERNEL_LT
+#define	TRSM_KERNEL_RN		STRSM_KERNEL_RN
+#define	TRSM_KERNEL_RT		STRSM_KERNEL_RT
+#define	TRSM_KERNEL_RR		STRSM_KERNEL_RN
+#define	TRSM_KERNEL_RC		STRSM_KERNEL_RT
+
+#define SYMM_IUTCOPY		SSYMM_IUTCOPY
+#define SYMM_ILTCOPY		SSYMM_ILTCOPY
+#define SYMM_OUTCOPY		SSYMM_OUTCOPY
+#define SYMM_OLTCOPY		SSYMM_OLTCOPY
+#define	TRMM_LNUU		STRMM_LNUU
+#define	TRMM_LNUN		STRMM_LNUN
+#define	TRMM_LNLU		STRMM_LNLU
+#define	TRMM_LNLN		STRMM_LNLN
+#define	TRMM_LTUU		STRMM_LTUU
+#define	TRMM_LTUN		STRMM_LTUN
+#define	TRMM_LTLU		STRMM_LTLU
+#define	TRMM_LTLN		STRMM_LTLN
+#define	TRMM_LRUU		STRMM_LNUU
+#define	TRMM_LRUN		STRMM_LNUN
+#define	TRMM_LRLU		STRMM_LNLU
+#define	TRMM_LRLN		STRMM_LNLN
+#define	TRMM_LCUU		STRMM_LTUU
+#define	TRMM_LCUN		STRMM_LTUN
+#define	TRMM_LCLU		STRMM_LTLU
+#define	TRMM_LCLN		STRMM_LTLN
+#define	TRMM_RNUU		STRMM_RNUU
+#define	TRMM_RNUN		STRMM_RNUN
+#define	TRMM_RNLU		STRMM_RNLU
+#define	TRMM_RNLN		STRMM_RNLN
+#define	TRMM_RTUU		STRMM_RTUU
+#define	TRMM_RTUN		STRMM_RTUN
+#define	TRMM_RTLU		STRMM_RTLU
+#define	TRMM_RTLN		STRMM_RTLN
+#define	TRMM_RRUU		STRMM_RNUU
+#define	TRMM_RRUN		STRMM_RNUN
+#define	TRMM_RRLU		STRMM_RNLU
+#define	TRMM_RRLN		STRMM_RNLN
+#define	TRMM_RCUU		STRMM_RTUU
+#define	TRMM_RCUN		STRMM_RTUN
+#define	TRMM_RCLU		STRMM_RTLU
+#define	TRMM_RCLN		STRMM_RTLN
+
+#define	TRSM_LNUU		STRSM_LNUU
+#define	TRSM_LNUN		STRSM_LNUN
+#define	TRSM_LNLU		STRSM_LNLU
+#define	TRSM_LNLN		STRSM_LNLN
+#define	TRSM_LTUU		STRSM_LTUU
+#define	TRSM_LTUN		STRSM_LTUN
+#define	TRSM_LTLU		STRSM_LTLU
+#define	TRSM_LTLN		STRSM_LTLN
+#define	TRSM_LRUU		STRSM_LNUU
+#define	TRSM_LRUN		STRSM_LNUN
+#define	TRSM_LRLU		STRSM_LNLU
+#define	TRSM_LRLN		STRSM_LNLN
+#define	TRSM_LCUU		STRSM_LTUU
+#define	TRSM_LCUN		STRSM_LTUN
+#define	TRSM_LCLU		STRSM_LTLU
+#define	TRSM_LCLN		STRSM_LTLN
+#define	TRSM_RNUU		STRSM_RNUU
+#define	TRSM_RNUN		STRSM_RNUN
+#define	TRSM_RNLU		STRSM_RNLU
+#define	TRSM_RNLN		STRSM_RNLN
+#define	TRSM_RTUU		STRSM_RTUU
+#define	TRSM_RTUN		STRSM_RTUN
+#define	TRSM_RTLU		STRSM_RTLU
+#define	TRSM_RTLN		STRSM_RTLN
+#define	TRSM_RRUU		STRSM_RNUU
+#define	TRSM_RRUN		STRSM_RNUN
+#define	TRSM_RRLU		STRSM_RNLU
+#define	TRSM_RRLN		STRSM_RNLN
+#define	TRSM_RCUU		STRSM_RTUU
+#define	TRSM_RCUN		STRSM_RTUN
+#define	TRSM_RCLU		STRSM_RTLU
+#define	TRSM_RCLN		STRSM_RTLN
+#define	SYRK_UN			SSYRK_UN
+#define	SYRK_UT			SSYRK_UT
+#define	SYRK_LN			SSYRK_LN
+#define	SYRK_LT			SSYRK_LT
+#define	SYRK_UR			SSYRK_UN
+#define	SYRK_UC			SSYRK_UT
+#define	SYRK_LR			SSYRK_LN
+#define	SYRK_LC			SSYRK_LT
+
+#define	SYRK_KERNEL_U		SSYRK_KERNEL_U
+#define	SYRK_KERNEL_L		SSYRK_KERNEL_L
+
+#define	HERK_UN			SSYRK_UN
+#define	HERK_LN			SSYRK_LN
+#define	HERK_UC			SSYRK_UT
+#define	HERK_LC			SSYRK_LT
+
+#define	HER2K_UN		SSYR2K_UN
+#define	HER2K_LN		SSYR2K_LN
+#define	HER2K_UC		SSYR2K_UT
+#define	HER2K_LC		SSYR2K_LT
+
+#define	SYR2K_UN		SSYR2K_UN
+#define	SYR2K_UT		SSYR2K_UT
+#define	SYR2K_LN		SSYR2K_LN
+#define	SYR2K_LT		SSYR2K_LT
+#define	SYR2K_UR		SSYR2K_UN
+#define	SYR2K_UC		SSYR2K_UT
+#define	SYR2K_LR		SSYR2K_LN
+#define	SYR2K_LC		SSYR2K_LT
+
+#define	SYR2K_KERNEL_U		SSYR2K_KERNEL_U
+#define	SYR2K_KERNEL_L		SSYR2K_KERNEL_L
+#define	SYRK_THREAD_UN		SSYRK_THREAD_UN
+#define	SYRK_THREAD_UT		SSYRK_THREAD_UT
+#define	SYRK_THREAD_LN		SSYRK_THREAD_LN
+#define	SYRK_THREAD_LT		SSYRK_THREAD_LT
+#define	SYRK_THREAD_UR		SSYRK_THREAD_UR
+#define	SYRK_THREAD_UC		SSYRK_THREAD_UC
+#define	SYRK_THREAD_LR		SSYRK_THREAD_LN
+#define	SYRK_THREAD_LC		SSYRK_THREAD_LT
+
+#define	HERK_THREAD_UN		SSYRK_THREAD_UN
+#define	HERK_THREAD_UT		SSYRK_THREAD_UT
+#define	HERK_THREAD_LN		SSYRK_THREAD_LN
+#define	HERK_THREAD_LT		SSYRK_THREAD_LT
+#define	HERK_THREAD_UR		SSYRK_THREAD_UR
+#define	HERK_THREAD_UC		SSYRK_THREAD_UC
+#define	HERK_THREAD_LR		SSYRK_THREAD_LN
+#define	HERK_THREAD_LC		SSYRK_THREAD_LT
+
+#define OMATCOPY_K_CN		SOMATCOPY_K_CN
+#define OMATCOPY_K_RN		SOMATCOPY_K_RN
+#define OMATCOPY_K_CT		SOMATCOPY_K_CT
+#define OMATCOPY_K_RT		SOMATCOPY_K_RT
+#define IMATCOPY_K_CN		SIMATCOPY_K_CN
+#define IMATCOPY_K_RN		SIMATCOPY_K_RN
+#define IMATCOPY_K_CT		SIMATCOPY_K_CT
+#define IMATCOPY_K_RT		SIMATCOPY_K_RT
+
+#define GEADD_K 		SGEADD_K
+
+#endif
+
 #else
 
 #define	AMAX_K			SAMAX_K
@@ -673,14 +956,14 @@
 #define	GEMV_S			SGEMV_S
 #define	GEMV_D			SGEMV_D
 
+
+#define	SYMV_U			SSYMV_U
+#define	SYMV_L			SSYMV_L
 #define	GERU_K			SGERU_K
 #define	GERC_K			SGERC_K
 #define	GERV_K			SGERV_K
 #define	GERD_K			SGERD_K
 
-#define	SYMV_U			SSYMV_U
-#define	SYMV_L			SSYMV_L
-
 #define	SYMV_THREAD_U		SSYMV_THREAD_U
 #define	SYMV_THREAD_L		SSYMV_THREAD_L
 
@@ -2202,6 +2485,9 @@
 #if defined(ARCH_X86) || defined(ARCH_X86_64) || defined(ARCH_IA64) || defined(ARCH_MIPS64) || defined(ARCH_ARM64)
 extern BLASLONG gemm_offset_a;
 extern BLASLONG gemm_offset_b;
+extern BLASLONG shgemm_p;
+extern BLASLONG shgemm_q;
+extern BLASLONG shgemm_r;
 extern BLASLONG sgemm_p;
 extern BLASLONG sgemm_q;
 extern BLASLONG sgemm_r;
diff --git a/common_mips.h b/common_mips.h
index 35bff5083..7dc3ba246 100644
--- a/common_mips.h
+++ b/common_mips.h
@@ -35,6 +35,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define MB  __sync_synchronize()
 #define WMB __sync_synchronize()
+#define RMB __sync_synchronize()
 
 #define INLINE inline
 
@@ -42,6 +43,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #ifndef ASSEMBLER
 
+#if !defined(MIPS24K)
 static inline unsigned int rpcc(void){
   unsigned long ret;
 
@@ -52,6 +54,7 @@ static inline unsigned int rpcc(void){
   return ret;
 }
 #define RPCC_DEFINED
+#endif
 
 static inline int blas_quickdivide(blasint x, blasint y){
   return x / y;
@@ -91,7 +94,7 @@ REALNAME:
 #endif
 #define HUGE_PAGESIZE   ( 4 << 20)
 
-#define BUFFER_SIZE     (16 << 20)
+#define BUFFER_SIZE     (16 << 21)
 
 
 #define BASE_ADDRESS (START_ADDRESS - BUFFER_SIZE * MAX_CPU_NUMBER)
diff --git a/common_mips64.h b/common_mips64.h
index 1163413dc..a06edfe08 100644
--- a/common_mips64.h
+++ b/common_mips64.h
@@ -73,6 +73,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define MB  __sync_synchronize()
 #define WMB __sync_synchronize()
+#define RMB __sync_synchronize()
 
 #define INLINE inline
 
@@ -226,7 +227,7 @@ REALNAME: ;\
 
 #define SEEK_ADDRESS
 
-#define BUFFER_SIZE     ( 32 << 20)
+#define BUFFER_SIZE     ( 32 << 21)
 
 #if defined(LOONGSON3A)
 #define PAGESIZE	(16UL << 10)
diff --git a/common_param.h b/common_param.h
index 574d5e176..c92609a76 100644
--- a/common_param.h
+++ b/common_param.h
@@ -47,6 +47,100 @@ typedef struct {
   int dtb_entries;
   int offsetA, offsetB, align;
 
+#ifdef BUILD_HALF
+  int shgemm_p, shgemm_q, shgemm_r;
+  int shgemm_unroll_m, shgemm_unroll_n, shgemm_unroll_mn;
+
+  float  (*shamax_k) (BLASLONG, float *, BLASLONG);
+  float  (*shamin_k) (BLASLONG, float *, BLASLONG);
+  float  (*shmax_k)  (BLASLONG, float *, BLASLONG);
+  float  (*shmin_k)  (BLASLONG, float *, BLASLONG);
+BLASLONG (*ishamax_k)(BLASLONG, float *, BLASLONG);
+BLASLONG (*ishamin_k)(BLASLONG, float *, BLASLONG);
+BLASLONG (*ishmax_k) (BLASLONG, float *, BLASLONG);
+BLASLONG (*ishmin_k) (BLASLONG, float *, BLASLONG);
+
+  float  (*shnrm2_k) (BLASLONG, float *, BLASLONG);
+  float  (*shasum_k) (BLASLONG, float *, BLASLONG);
+  float  (*shsum_k)  (BLASLONG, float *, BLASLONG);
+  int    (*shcopy_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);
+  float  (*shdot_k)  (BLASLONG, float *, BLASLONG, float *, BLASLONG);
+  double (*dshdot_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);
+
+  int    (*shrot_k)  (BLASLONG, float *, BLASLONG, float *, BLASLONG, float, float);
+
+  int    (*shaxpy_k) (BLASLONG, BLASLONG, BLASLONG, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG);
+  int    (*shscal_k) (BLASLONG, BLASLONG, BLASLONG, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG);
+  int    (*shswap_k) (BLASLONG, BLASLONG, BLASLONG, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG);
+
+  int    (*shgemv_n) (BLASLONG, BLASLONG, BLASLONG, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG, float *);
+  int    (*shgemv_t) (BLASLONG, BLASLONG, BLASLONG, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG, float *);
+  int    (*shger_k)  (BLASLONG, BLASLONG, BLASLONG, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG, float *);
+
+  int    (*shsymv_L) (BLASLONG, BLASLONG, float,  float  *, BLASLONG, float  *, BLASLONG, float  *, BLASLONG, float *);
+  int    (*shsymv_U) (BLASLONG, BLASLONG, float,  float  *, BLASLONG, float  *, BLASLONG, float  *, BLASLONG, float *);
+
+  int    (*shgemm_kernel   )(BLASLONG, BLASLONG, BLASLONG, float, bfloat16 *, bfloat16 *, float *, BLASLONG);
+  int    (*shgemm_beta     )(BLASLONG, BLASLONG, BLASLONG, float, bfloat16 *, BLASLONG, bfloat16 *, BLASLONG, float *, BLASLONG);
+
+  int    (*shgemm_incopy   )(BLASLONG, BLASLONG, bfloat16 *, BLASLONG, bfloat16 *);
+  int    (*shgemm_itcopy   )(BLASLONG, BLASLONG, bfloat16 *, BLASLONG, bfloat16 *);
+  int    (*shgemm_oncopy   )(BLASLONG, BLASLONG, bfloat16 *, BLASLONG, bfloat16 *);
+  int    (*shgemm_otcopy   )(BLASLONG, BLASLONG, bfloat16 *, BLASLONG, bfloat16 *);
+
+  int    (*shtrsm_kernel_LN)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
+  int    (*shtrsm_kernel_LT)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
+  int    (*shtrsm_kernel_RN)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
+  int    (*shtrsm_kernel_RT)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
+
+  int    (*shtrsm_iunucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
+  int    (*shtrsm_iunncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
+  int    (*shtrsm_iutucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
+  int    (*shtrsm_iutncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
+  int    (*shtrsm_ilnucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
+  int    (*shtrsm_ilnncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
+  int    (*shtrsm_iltucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
+  int    (*shtrsm_iltncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
+  int    (*shtrsm_ounucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
+  int    (*shtrsm_ounncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
+  int    (*shtrsm_outucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
+  int    (*shtrsm_outncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
+  int    (*shtrsm_olnucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
+  int    (*shtrsm_olnncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
+  int    (*shtrsm_oltucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
+  int    (*shtrsm_oltncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, float *);
+
+  int    (*shtrmm_kernel_RN)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
+  int    (*shtrmm_kernel_RT)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
+  int    (*shtrmm_kernel_LN)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
+  int    (*shtrmm_kernel_LT)(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG, BLASLONG);
+
+  int    (*shtrmm_iunucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shtrmm_iunncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shtrmm_iutucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shtrmm_iutncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shtrmm_ilnucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shtrmm_ilnncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shtrmm_iltucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shtrmm_iltncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shtrmm_ounucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shtrmm_ounncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shtrmm_outucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shtrmm_outncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shtrmm_olnucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shtrmm_olnncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shtrmm_oltucopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shtrmm_oltncopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+
+  int    (*shsymm_iutcopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shsymm_iltcopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shsymm_outcopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+  int    (*shsymm_oltcopy)(BLASLONG, BLASLONG, float *, BLASLONG, BLASLONG, BLASLONG, float *);
+
+  int	 (*shneg_tcopy)   (BLASLONG, BLASLONG, float *, BLASLONG, float *);
+  int    (*shlaswp_ncopy) (BLASLONG, BLASLONG, BLASLONG, float *, BLASLONG, blasint *, float *);
+
+#endif
   int sgemm_p, sgemm_q, sgemm_r;
   int sgemm_unroll_m, sgemm_unroll_n, sgemm_unroll_mn;
 
@@ -84,6 +178,7 @@ BLASLONG (*ismin_k) (BLASLONG, float *, BLASLONG);
   int    (*sgemm_kernel   )(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG);
   int    (*sgemm_beta     )(BLASLONG, BLASLONG, BLASLONG, float, float *, BLASLONG, float *, BLASLONG, float  *, BLASLONG);
 
+
   int    (*sgemm_incopy   )(BLASLONG, BLASLONG, float *, BLASLONG, float *);
   int    (*sgemm_itcopy   )(BLASLONG, BLASLONG, float *, BLASLONG, float *);
   int    (*sgemm_oncopy   )(BLASLONG, BLASLONG, float *, BLASLONG, float *);
@@ -907,6 +1002,15 @@ extern gotoblas_t *gotoblas;
 
 #define HAVE_EX_L2	gotoblas -> exclusive_cache
 
+#ifdef BUILD_HALF
+#define	SHGEMM_P		gotoblas -> shgemm_p
+#define	SHGEMM_Q		gotoblas -> shgemm_q
+#define	SHGEMM_R		gotoblas -> shgemm_r
+#define	SHGEMM_UNROLL_M	gotoblas -> shgemm_unroll_m
+#define	SHGEMM_UNROLL_N	gotoblas -> shgemm_unroll_n
+#define	SHGEMM_UNROLL_MN	gotoblas -> shgemm_unroll_mn
+#endif
+
 #define	SGEMM_P		gotoblas -> sgemm_p
 #define	SGEMM_Q		gotoblas -> sgemm_q
 #define	SGEMM_R		gotoblas -> sgemm_r
@@ -984,6 +1088,19 @@ extern gotoblas_t *gotoblas;
 #define HAVE_EX_L2	0
 #endif
 
+#ifdef BUILD_HALF
+#define	SHGEMM_P		SHGEMM_DEFAULT_P
+#define	SHGEMM_Q		SHGEMM_DEFAULT_Q
+#define	SHGEMM_R		SHGEMM_DEFAULT_R
+#define SHGEMM_UNROLL_M	SHGEMM_DEFAULT_UNROLL_M
+#define SHGEMM_UNROLL_N	SHGEMM_DEFAULT_UNROLL_N
+#ifdef  SHGEMM_DEFAULT_UNROLL_MN
+#define SHGEMM_UNROLL_MN	SHGEMM_DEFAULT_UNROLL_MN
+#else
+#define SHGEMM_UNROLL_MN	MAX((SHGEMM_UNROLL_M), (SHGEMM_UNROLL_N))
+#endif
+#endif
+
 #define	SGEMM_P		SGEMM_DEFAULT_P
 #define	SGEMM_Q		SGEMM_DEFAULT_Q
 #define	SGEMM_R		SGEMM_DEFAULT_R
@@ -1119,6 +1236,18 @@ extern gotoblas_t *gotoblas;
 #define GEMM_DEFAULT_R		DGEMM_DEFAULT_R
 #define GEMM_DEFAULT_UNROLL_M	DGEMM_DEFAULT_UNROLL_M
 #define GEMM_DEFAULT_UNROLL_N	DGEMM_DEFAULT_UNROLL_N
+#elif defined(HALF)
+#define GEMM_P			SHGEMM_P
+#define GEMM_Q			SHGEMM_Q
+#define GEMM_R			SHGEMM_R
+#define GEMM_UNROLL_M		SHGEMM_UNROLL_M
+#define GEMM_UNROLL_N		SHGEMM_UNROLL_N
+#define GEMM_UNROLL_MN		SHGEMM_UNROLL_MN
+#define GEMM_DEFAULT_P		SHGEMM_DEFAULT_P
+#define GEMM_DEFAULT_Q		SHGEMM_DEFAULT_Q
+#define GEMM_DEFAULT_R		SHGEMM_DEFAULT_R
+#define GEMM_DEFAULT_UNROLL_M	SHGEMM_DEFAULT_UNROLL_M
+#define GEMM_DEFAULT_UNROLL_N	SHGEMM_DEFAULT_UNROLL_N
 #else
 #define GEMM_P			SGEMM_P
 #define GEMM_Q			SGEMM_Q
@@ -1204,28 +1333,32 @@ extern gotoblas_t *gotoblas;
 #define GEMM_THREAD gemm_thread_n
 #endif
 
+#ifndef SHGEMM_DEFAULT_R
+#define SHGEMM_DEFAULT_R (((BUFFER_SIZE - ((SHGEMM_DEFAULT_P * SHGEMM_DEFAULT_Q *  4 + GEMM_DEFAULT_OFFSET_A + GEMM_DEFAULT_ALIGN) & ~GEMM_DEFAULT_ALIGN)) / (SHGEMM_DEFAULT_Q *  4) - 15) & ~15UL)
+#endif
+
 #ifndef SGEMM_DEFAULT_R
-#define SGEMM_DEFAULT_R (((BUFFER_SIZE - ((SGEMM_DEFAULT_P * SGEMM_DEFAULT_Q *  4 + GEMM_DEFAULT_OFFSET_A + GEMM_DEFAULT_ALIGN) & ~GEMM_DEFAULT_ALIGN)) / (SGEMM_DEFAULT_Q *  4) - 15) & ~15)
+#define SGEMM_DEFAULT_R (((BUFFER_SIZE - ((SGEMM_DEFAULT_P * SGEMM_DEFAULT_Q *  4 + GEMM_DEFAULT_OFFSET_A + GEMM_DEFAULT_ALIGN) & ~GEMM_DEFAULT_ALIGN)) / (SGEMM_DEFAULT_Q *  4) - 15) & ~15UL)
 #endif
 
 #ifndef DGEMM_DEFAULT_R
-#define DGEMM_DEFAULT_R (((BUFFER_SIZE - ((DGEMM_DEFAULT_P * DGEMM_DEFAULT_Q *  8 + GEMM_DEFAULT_OFFSET_A + GEMM_DEFAULT_ALIGN) & ~GEMM_DEFAULT_ALIGN)) / (DGEMM_DEFAULT_Q *  8) - 15) & ~15)
+#define DGEMM_DEFAULT_R (((BUFFER_SIZE - ((DGEMM_DEFAULT_P * DGEMM_DEFAULT_Q *  8 + GEMM_DEFAULT_OFFSET_A + GEMM_DEFAULT_ALIGN) & ~GEMM_DEFAULT_ALIGN)) / (DGEMM_DEFAULT_Q *  8) - 15) & ~15UL)
 #endif
 
 #ifndef QGEMM_DEFAULT_R
-#define QGEMM_DEFAULT_R (((BUFFER_SIZE - ((QGEMM_DEFAULT_P * QGEMM_DEFAULT_Q * 16 + GEMM_DEFAULT_OFFSET_A + GEMM_DEFAULT_ALIGN) & ~GEMM_DEFAULT_ALIGN)) / (QGEMM_DEFAULT_Q * 16) - 15) & ~15)
+#define QGEMM_DEFAULT_R (((BUFFER_SIZE - ((QGEMM_DEFAULT_P * QGEMM_DEFAULT_Q * 16 + GEMM_DEFAULT_OFFSET_A + GEMM_DEFAULT_ALIGN) & ~GEMM_DEFAULT_ALIGN)) / (QGEMM_DEFAULT_Q * 16) - 15) & ~15UL)
 #endif
 
 #ifndef CGEMM_DEFAULT_R
-#define CGEMM_DEFAULT_R (((BUFFER_SIZE - ((CGEMM_DEFAULT_P * CGEMM_DEFAULT_Q *  8 + GEMM_DEFAULT_OFFSET_A + GEMM_DEFAULT_ALIGN) & ~GEMM_DEFAULT_ALIGN)) / (CGEMM_DEFAULT_Q *  8) - 15) & ~15)
+#define CGEMM_DEFAULT_R (((BUFFER_SIZE - ((CGEMM_DEFAULT_P * CGEMM_DEFAULT_Q *  8 + GEMM_DEFAULT_OFFSET_A + GEMM_DEFAULT_ALIGN) & ~GEMM_DEFAULT_ALIGN)) / (CGEMM_DEFAULT_Q *  8) - 15) & ~15UL)
 #endif
 
 #ifndef ZGEMM_DEFAULT_R
-#define ZGEMM_DEFAULT_R (((BUFFER_SIZE - ((ZGEMM_DEFAULT_P * ZGEMM_DEFAULT_Q * 16 + GEMM_DEFAULT_OFFSET_A + GEMM_DEFAULT_ALIGN) & ~GEMM_DEFAULT_ALIGN)) / (ZGEMM_DEFAULT_Q * 16) - 15) & ~15)
+#define ZGEMM_DEFAULT_R (((BUFFER_SIZE - ((ZGEMM_DEFAULT_P * ZGEMM_DEFAULT_Q * 16 + GEMM_DEFAULT_OFFSET_A + GEMM_DEFAULT_ALIGN) & ~GEMM_DEFAULT_ALIGN)) / (ZGEMM_DEFAULT_Q * 16) - 15) & ~15UL)
 #endif
 
 #ifndef XGEMM_DEFAULT_R
-#define XGEMM_DEFAULT_R (((BUFFER_SIZE - ((XGEMM_DEFAULT_P * XGEMM_DEFAULT_Q * 32 + GEMM_DEFAULT_OFFSET_A + GEMM_DEFAULT_ALIGN) & ~GEMM_DEFAULT_ALIGN)) / (XGEMM_DEFAULT_Q * 32) - 15) & ~15)
+#define XGEMM_DEFAULT_R (((BUFFER_SIZE - ((XGEMM_DEFAULT_P * XGEMM_DEFAULT_Q * 32 + GEMM_DEFAULT_OFFSET_A + GEMM_DEFAULT_ALIGN) & ~GEMM_DEFAULT_ALIGN)) / (XGEMM_DEFAULT_Q * 32) - 15) & ~15UL)
 #endif
 
 #ifndef SNUMOPT
diff --git a/common_power.h b/common_power.h
index e7caf9adf..e29d0f382 100644
--- a/common_power.h
+++ b/common_power.h
@@ -71,9 +71,11 @@
 #if defined(POWER8) || defined(POWER9)
 #define MB		__asm__ __volatile__ ("eieio":::"memory")
 #define WMB		__asm__ __volatile__ ("eieio":::"memory")
+#define RMB		__asm__ __volatile__ ("eieio":::"memory")
 #else
 #define MB		__asm__ __volatile__ ("sync")
 #define WMB		__asm__ __volatile__ ("sync")
+#define RMB		__asm__ __volatile__ ("sync")
 #endif
 
 #define INLINE inline
diff --git a/common_sh.h b/common_sh.h
new file mode 100644
index 000000000..7a0045762
--- /dev/null
+++ b/common_sh.h
@@ -0,0 +1,65 @@
+#ifndef COMMON_SH_H
+#define COMMON_SH_H
+
+#ifndef DYNAMIC_ARCH
+
+#define	SHGEMM_ONCOPY		shgemm_oncopy
+#define	SHGEMM_OTCOPY		shgemm_otcopy
+
+#if SHGEMM_DEFAULT_UNROLL_M == SHGEMM_DEFAULT_UNROLL_N
+#define	SHGEMM_INCOPY		shgemm_oncopy
+#define	SHGEMM_ITCOPY		shgemm_otcopy
+#else
+#define	SHGEMM_INCOPY		shgemm_incopy
+#define	SHGEMM_ITCOPY		shgemm_itcopy
+#endif
+#define	SHGEMM_BETA		shgemm_beta
+#define SHGEMM_KERNEL            shgemm_kernel
+
+#else
+
+#define	SHGEMM_ONCOPY		gotoblas -> shgemm_oncopy
+#define	SHGEMM_OTCOPY		gotoblas -> shgemm_otcopy
+#define	SHGEMM_INCOPY		gotoblas -> shgemm_incopy
+#define	SHGEMM_ITCOPY		gotoblas -> shgemm_itcopy
+#define	SHGEMM_BETA		gotoblas -> shgemm_beta
+#define	SHGEMM_KERNEL		gotoblas -> shgemm_kernel
+
+#endif
+
+#define	SHGEMM_NN		shgemm_nn
+#define	SHGEMM_CN		shgemm_tn
+#define	SHGEMM_TN		shgemm_tn
+#define	SHGEMM_NC		shgemm_nt
+#define	SHGEMM_NT		shgemm_nt
+#define	SHGEMM_CC		shgemm_tt
+#define	SHGEMM_CT		shgemm_tt
+#define	SHGEMM_TC		shgemm_tt
+#define	SHGEMM_TT		shgemm_tt
+#define	SHGEMM_NR		shgemm_nn
+#define	SHGEMM_TR		shgemm_tn
+#define	SHGEMM_CR		shgemm_tn
+#define	SHGEMM_RN		shgemm_nn
+#define	SHGEMM_RT		shgemm_nt
+#define	SHGEMM_RC		shgemm_nt
+#define	SHGEMM_RR		shgemm_nn
+
+#define	SHGEMM_THREAD_NN		shgemm_thread_nn
+#define	SHGEMM_THREAD_CN		shgemm_thread_tn
+#define	SHGEMM_THREAD_TN		shgemm_thread_tn
+#define	SHGEMM_THREAD_NC		shgemm_thread_nt
+#define	SHGEMM_THREAD_NT		shgemm_thread_nt
+#define	SHGEMM_THREAD_CC		shgemm_thread_tt
+#define	SHGEMM_THREAD_CT		shgemm_thread_tt
+#define	SHGEMM_THREAD_TC		shgemm_thread_tt
+#define	SHGEMM_THREAD_TT		shgemm_thread_tt
+#define	SHGEMM_THREAD_NR		shgemm_thread_nn
+#define	SHGEMM_THREAD_TR		shgemm_thread_tn
+#define	SHGEMM_THREAD_CR		shgemm_thread_tn
+#define	SHGEMM_THREAD_RN		shgemm_thread_nn
+#define	SHGEMM_THREAD_RT		shgemm_thread_nt
+#define	SHGEMM_THREAD_RC		shgemm_thread_nt
+#define	SHGEMM_THREAD_RR		shgemm_thread_nn
+
+#endif
+
diff --git a/common_sparc.h b/common_sparc.h
index f99972db9..85e29fffa 100644
--- a/common_sparc.h
+++ b/common_sparc.h
@@ -41,6 +41,7 @@
 
 #define MB	__asm__ __volatile__ ("nop")
 #define WMB	__asm__ __volatile__ ("nop")
+#define RMB	__asm__ __volatile__ ("nop")
 
 #ifndef ASSEMBLER
 
diff --git a/common_x86.h b/common_x86.h
index 99adc9f5b..ec928e236 100644
--- a/common_x86.h
+++ b/common_x86.h
@@ -47,6 +47,7 @@
 
 #define MB
 #define WMB
+#define RMB
 
 #ifdef C_SUN
 #define	__asm__ __asm
diff --git a/common_x86_64.h b/common_x86_64.h
index fe5539abe..15d0c30aa 100644
--- a/common_x86_64.h
+++ b/common_x86_64.h
@@ -63,13 +63,16 @@
 #ifdef __GNUC__
 #define MB do { __asm__ __volatile__("": : :"memory"); } while (0)
 #define WMB do { __asm__ __volatile__("": : :"memory"); } while (0)
+#define RMB
 #else
 #define MB do {} while (0)
 #define WMB do {} while (0)
+#define RMB
 #endif
 
 static void __inline blas_lock(volatile BLASULONG *address){
 
+	
 #ifndef C_MSVC
   int ret;
 #else
@@ -77,7 +80,7 @@ static void __inline blas_lock(volatile BLASULONG *address){
 #endif
 
   do {
-    while (*address) {YIELDING;};
+    while (*address) {YIELDING;}
 
 #ifndef C_MSVC
     __asm__ __volatile__(
@@ -196,9 +199,9 @@ static __inline BLASLONG blas_quickdivide(BLASLONG x, BLASLONG y){
 #else
 extern unsigned int blas_quick_divide_table[];
 
-static __inline int blas_quickdivide(unsigned int x, unsigned int y){
+static __inline unsigned int blas_quickdivide(unsigned int x, unsigned int y){
 
-  unsigned int result;
+  volatile unsigned int result;
 
   if (y <= 1) return x;
 
@@ -212,7 +215,6 @@ static __inline int blas_quickdivide(unsigned int x, unsigned int y){
   y = blas_quick_divide_table[y];
 
   __asm__ __volatile__  ("mull %0" :"=d" (result), "+a"(x) : "0" (y));
-
   return result;
 }
 #endif
@@ -226,7 +228,13 @@ static __inline int blas_quickdivide(unsigned int x, unsigned int y){
 #define HUGE_PAGESIZE	( 2 << 20)
 
 #ifndef BUFFERSIZE
+#if defined(SKYLAKEX) 
+#define BUFFER_SIZE	(32 << 21)
+#elif defined(HASWELL) || defined(ZEN)
+#define BUFFER_SIZE	(32 << 22)
+#else
 #define BUFFER_SIZE	(32 << 20)
+#endif
 #else
 #define BUFFER_SIZE	(32 << BUFFERSIZE)
 #endif
diff --git a/common_zarch.h b/common_zarch.h
index e105574e0..442bae821 100644
--- a/common_zarch.h
+++ b/common_zarch.h
@@ -34,9 +34,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define COMMON_ZARCH
 
 #define MB   
-//__asm__ __volatile__ ("dmb  ish" : : : "memory")
 #define WMB  
-//__asm__ __volatile__ ("dmb  ishst" : : : "memory")
+#define RMB
 
 
 #define INLINE inline
@@ -123,11 +122,7 @@ REALNAME:
 #endif
 #define HUGE_PAGESIZE   ( 4 << 20)
 
-#if defined(CORTEXA57)
-#define BUFFER_SIZE     (20 << 20)
-#else
-#define BUFFER_SIZE     (16 << 20)
-#endif
+#define BUFFER_SIZE     (32 << 22)
 
 
 #define BASE_ADDRESS (START_ADDRESS - BUFFER_SIZE * MAX_CPU_NUMBER)
diff --git a/cpp_thread_test/dgemm_thread_safety.cpp b/cpp_thread_test/dgemm_thread_safety.cpp
index cecf794fa..1c5287524 100644
--- a/cpp_thread_test/dgemm_thread_safety.cpp
+++ b/cpp_thread_test/dgemm_thread_safety.cpp
@@ -12,9 +12,13 @@ void launch_cblas_dgemm(double* A, double* B, double* C, const blasint randomMat
 
 int main(int argc, char* argv[]){
 	blasint randomMatSize = 1024; //dimension of the random square matrices used
-	uint32_t numConcurrentThreads = 52; //number of concurrent calls of the functions being tested
+	uint32_t numConcurrentThreads = 96; //number of concurrent calls of the functions being tested
 	uint32_t numTestRounds = 16; //number of testing rounds before success exit
+	uint32_t maxHwThreads = omp_get_max_threads();
 	
+	if (maxHwThreads < 96)
+		numConcurrentThreads = maxHwThreads;
+
 	if (argc > 4){
 		std::cout<<"ERROR: too many arguments for thread safety tester"<<std::endl;
 		abort();
diff --git a/cpp_thread_test/dgemv_thread_safety.cpp b/cpp_thread_test/dgemv_thread_safety.cpp
index 22505d03f..5411fec29 100644
--- a/cpp_thread_test/dgemv_thread_safety.cpp
+++ b/cpp_thread_test/dgemv_thread_safety.cpp
@@ -15,6 +15,10 @@ int main(int argc, char* argv[]){
 	blasint randomMatSize = 1024; //dimension of the random square matrices and vectors being used
 	uint32_t numConcurrentThreads = 52; //number of concurrent calls of the functions being tested
 	uint32_t numTestRounds = 16; //number of testing rounds before success exit
+	uint32_t maxHwThreads = omp_get_max_threads();
+	
+	if (maxHwThreads < 52)
+		numConcurrentThreads = maxHwThreads -4;
 	
 	if (argc > 4){
 		std::cout<<"ERROR: too many arguments for thread safety tester"<<std::endl;
diff --git a/cpuid_mips.c b/cpuid_mips.c
index 6f2932c94..3a2e12393 100644
--- a/cpuid_mips.c
+++ b/cpuid_mips.c
@@ -73,11 +73,13 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define CPU_UNKNOWN     0
 #define CPU_P5600       1
 #define CPU_1004K	2
+#define CPU_24K		3
 
 static char *cpuname[] = {
   "UNKNOWN",
   "P5600",
-  "1004K"
+  "MIPS1004K",
+  "MIPS24K"
 };
 
 int detect(void){
@@ -105,6 +107,8 @@ int detect(void){
     return CPU_P5600;
   } else if (strstr(p, "1004K")) {
     return CPU_1004K;
+  } else if (strstr(p, " 24K")) {
+    return CPU_24K;
   } else  
     return CPU_UNKNOWN;
   }
@@ -121,7 +125,7 @@ void get_architecture(void){
 }
 
 void get_subarchitecture(void){
-  if(detect()==CPU_P5600|| detect()==CPU_1004K){
+  if(detect()==CPU_P5600|| detect()==CPU_1004K|| detect()==CPU_24K){
     printf("P5600");
   }else{
     printf("UNKNOWN");
@@ -146,7 +150,15 @@ void get_cpuconfig(void){
     printf("#define MIPS1004K\n");
     printf("#define L1_DATA_SIZE 32768\n");
     printf("#define L1_DATA_LINESIZE 32\n");
-    printf("#define L2_SIZE 26144\n");
+    printf("#define L2_SIZE 262144\n");
+    printf("#define DTB_DEFAULT_ENTRIES 8\n");
+    printf("#define DTB_SIZE 4096\n");
+    printf("#define L2_ASSOCIATIVE 4\n");
+  } else if (detect()==CPU_24K) {
+    printf("#define MIPS24K\n");
+    printf("#define L1_DATA_SIZE 32768\n");
+    printf("#define L1_DATA_LINESIZE 32\n");
+    printf("#define L2_SIZE 32768\n");
     printf("#define DTB_DEFAULT_ENTRIES 8\n");
     printf("#define DTB_SIZE 4096\n");
     printf("#define L2_ASSOCIATIVE 4\n");
@@ -159,7 +171,9 @@ void get_libname(void){
   if(detect()==CPU_P5600) {
     printf("p5600\n");
   } else if (detect()==CPU_1004K) {
-    printf("1004K\n");
+    printf("mips1004K\n");
+  } else if (detect()==CPU_24K) {
+    printf("mips24K\n");
   }else{
     printf("mips\n");
   }
diff --git a/ctest/CMakeLists.txt b/ctest/CMakeLists.txt
index 14c9d1944..8d301c239 100644
--- a/ctest/CMakeLists.txt
+++ b/ctest/CMakeLists.txt
@@ -12,6 +12,9 @@ FILE(WRITE ${CMAKE_CURRENT_BINARY_DIR}/test_cblas_helper.sh
 foreach(float_type ${FLOAT_TYPES})
   string(SUBSTRING ${float_type} 0 1 float_char_upper)
   string(TOLOWER ${float_char_upper} float_char)
+  if (${float_char} STREQUAL "h")
+   continue()
+  endif()
   #level1
   add_executable(x${float_char}cblat1
     c_${float_char}blat1.f
diff --git a/ctest/Makefile b/ctest/Makefile
index f562c9bb3..6f5b65142 100644
--- a/ctest/Makefile
+++ b/ctest/Makefile
@@ -40,8 +40,11 @@ ztestl3o = c_zblas3.o c_z3chke.o auxiliary.o c_xerbla.o constant.o
 ztestl3o_3m = c_zblas3_3m.o c_z3chke_3m.o auxiliary.o c_xerbla.o constant.o
 
 
-
+ifeq ($(NOFORTRAN),1)
+all ::
+else
 all :: all1 all2 all3
+endif
 
 all1: xscblat1 xdcblat1 xccblat1 xzcblat1
 ifndef CROSS
diff --git a/ctest/c_dblas1.c b/ctest/c_dblas1.c
index a288154c2..e49ae6007 100644
--- a/ctest/c_dblas1.c
+++ b/ctest/c_dblas1.c
@@ -53,6 +53,13 @@ void F77_drot( const int *N, double *X, const int *incX, double *Y,
    return;
 }
 
+void F77_drotm(const int *N, double *X, const int *incX, double *Y,
+               const int *incY, const double *dparam)
+{
+   cblas_drotm(*N, X, *incX, Y, *incY, dparam);
+   return;
+}
+
 void F77_dscal(const int *N, const double *alpha, double *X,
                          const int *incX)
 {
diff --git a/ctest/c_dblat1.f b/ctest/c_dblat1.f
index 4a71b4dcf..0139ede63 100644
--- a/ctest/c_dblat1.f
+++ b/ctest/c_dblat1.f
@@ -19,7 +19,7 @@
       DATA             SFAC/9.765625D-4/
 *     .. Executable Statements ..
       WRITE (NOUT,99999)
-      DO 20 IC = 1, 10
+      DO 20 IC = 1, 11
          ICASE = IC
          CALL HEADER
 *
@@ -40,7 +40,7 @@
          ELSE IF (ICASE.EQ.1 .OR. ICASE.EQ.2 .OR. ICASE.EQ.5 .OR.
      +            ICASE.EQ.6) THEN
             CALL CHECK2(SFAC)
-         ELSE IF (ICASE.EQ.4) THEN
+         ELSE IF (ICASE.EQ.4 .OR. ICASE.EQ.11) THEN
             CALL CHECK3(SFAC)
          END IF
 *        -- Print
@@ -59,7 +59,7 @@
       INTEGER          ICASE, INCX, INCY, MODE, N
       LOGICAL          PASS
 *     .. Local Arrays ..
-      CHARACTER*15      L(10)
+      CHARACTER*15      L(11)
 *     .. Common blocks ..
       COMMON           /COMBLA/ICASE, N, INCX, INCY, MODE, PASS
 *     .. Data statements ..
@@ -73,6 +73,7 @@
       DATA             L(8)/'CBLAS_DASUM '/
       DATA             L(9)/'CBLAS_DSCAL '/
       DATA             L(10)/'CBLAS_IDAMAX'/
+      DATA             L(11)/'CBLAS_DROTM'/
 *     .. Executable Statements ..
       WRITE (NOUT,99999) ICASE, L(ICASE)
       RETURN
@@ -400,199 +401,81 @@
       LOGICAL           PASS
 *     .. Local Scalars ..
       DOUBLE PRECISION  SC, SS
-      INTEGER           I, K, KI, KN, KSIZE, LENX, LENY, MX, MY
+      INTEGER           I, KI, KN, KSIZE, LEN
 *     .. Local Arrays ..
-      DOUBLE PRECISION  COPYX(5), COPYY(5), DT9X(7,4,4), DT9Y(7,4,4),
-     +                  DX1(7), DY1(7), MWPC(11), MWPS(11), MWPSTX(5),
-     +                  MWPSTY(5), MWPTX(11,5), MWPTY(11,5), MWPX(5),
-     +                  MWPY(5), SSIZE2(14,2), STX(7), STY(7), SX(7),
-     +                  SY(7)
-      INTEGER           INCXS(4), INCYS(4), LENS(4,2), MWPINX(11),
-     +                  MWPINY(11), MWPN(11), NS(4)
+      DOUBLE PRECISION  DX(10), DY(10), SSIZE2(10,2), STX(10), 
+     +                  STY(10), SX(10), SY(10),
+     +                  PARAM(5, 4), DPARAM(5) 
+      INTEGER           INCXS(7), INCYS(7), NS(5)
 *     .. External Subroutines ..
-      EXTERNAL          STEST,DROTTEST
+      EXTERNAL          STEST, DROTTEST, DROT
 *     .. Intrinsic Functions ..
-      INTRINSIC         ABS, MIN
+      INTRINSIC         MIN
 *     .. Common blocks ..
       COMMON            /COMBLA/ICASE, N, INCX, INCY, MODE, PASS
 *     .. Data statements ..
-      DATA              INCXS/1, 2, -2, -1/
-      DATA              INCYS/1, -2, 1, -2/
-      DATA              LENS/1, 1, 2, 4, 1, 1, 3, 7/
-      DATA              NS/0, 1, 2, 4/
-      DATA              DX1/0.6D0, 0.1D0, -0.5D0, 0.8D0, 0.9D0, -0.3D0,
-     +                  -0.4D0/
-      DATA              DY1/0.5D0, -0.9D0, 0.3D0, 0.7D0, -0.6D0, 0.2D0,
-     +                  0.8D0/
+      DATA              INCXS/1, 1, 2, 2, -2, -1, -2/
+      DATA              INCYS/1, 2, 2, -2, 1, -2, -2/
+      DATA              NS/0, 1, 2, 4, 5/
+      DATA              DX/0.6D0, 0.1D0, -0.5D0, 0.8D0, 0.9D0, -0.3D0,
+     +                  -0.4D0, 0.7D0, 0.5D0, 0.2D0/
+      DATA              DY/0.5D0, -0.9D0, 0.3D0, 0.7D0, -0.6D0, 0.2D0,
+     +                  0.8D0, -0.5D0, 0.1D0, -0.3D0/
       DATA              SC, SS/0.8D0, 0.6D0/
-      DATA              DT9X/0.6D0, 0.0D0, 0.0D0, 0.0D0, 0.0D0, 0.0D0,
-     +                  0.0D0, 0.78D0, 0.0D0, 0.0D0, 0.0D0, 0.0D0,
-     +                  0.0D0, 0.0D0, 0.78D0, -0.46D0, 0.0D0, 0.0D0,
-     +                  0.0D0, 0.0D0, 0.0D0, 0.78D0, -0.46D0, -0.22D0,
-     +                  1.06D0, 0.0D0, 0.0D0, 0.0D0, 0.6D0, 0.0D0,
-     +                  0.0D0, 0.0D0, 0.0D0, 0.0D0, 0.0D0, 0.78D0,
-     +                  0.0D0, 0.0D0, 0.0D0, 0.0D0, 0.0D0, 0.0D0,
-     +                  0.66D0, 0.1D0, -0.1D0, 0.0D0, 0.0D0, 0.0D0,
-     +                  0.0D0, 0.96D0, 0.1D0, -0.76D0, 0.8D0, 0.90D0,
-     +                  -0.3D0, -0.02D0, 0.6D0, 0.0D0, 0.0D0, 0.0D0,
-     +                  0.0D0, 0.0D0, 0.0D0, 0.78D0, 0.0D0, 0.0D0,
-     +                  0.0D0, 0.0D0, 0.0D0, 0.0D0, -0.06D0, 0.1D0,
-     +                  -0.1D0, 0.0D0, 0.0D0, 0.0D0, 0.0D0, 0.90D0,
-     +                  0.1D0, -0.22D0, 0.8D0, 0.18D0, -0.3D0, -0.02D0,
-     +                  0.6D0, 0.0D0, 0.0D0, 0.0D0, 0.0D0, 0.0D0, 0.0D0,
-     +                  0.78D0, 0.0D0, 0.0D0, 0.0D0, 0.0D0, 0.0D0,
-     +                  0.0D0, 0.78D0, 0.26D0, 0.0D0, 0.0D0, 0.0D0,
-     +                  0.0D0, 0.0D0, 0.78D0, 0.26D0, -0.76D0, 1.12D0,
-     +                  0.0D0, 0.0D0, 0.0D0/
-      DATA              DT9Y/0.5D0, 0.0D0, 0.0D0, 0.0D0, 0.0D0, 0.0D0,
-     +                  0.0D0, 0.04D0, 0.0D0, 0.0D0, 0.0D0, 0.0D0,
-     +                  0.0D0, 0.0D0, 0.04D0, -0.78D0, 0.0D0, 0.0D0,
-     +                  0.0D0, 0.0D0, 0.0D0, 0.04D0, -0.78D0, 0.54D0,
-     +                  0.08D0, 0.0D0, 0.0D0, 0.0D0, 0.5D0, 0.0D0,
-     +                  0.0D0, 0.0D0, 0.0D0, 0.0D0, 0.0D0, 0.04D0,
-     +                  0.0D0, 0.0D0, 0.0D0, 0.0D0, 0.0D0, 0.0D0, 0.7D0,
-     +                  -0.9D0, -0.12D0, 0.0D0, 0.0D0, 0.0D0, 0.0D0,
-     +                  0.64D0, -0.9D0, -0.30D0, 0.7D0, -0.18D0, 0.2D0,
-     +                  0.28D0, 0.5D0, 0.0D0, 0.0D0, 0.0D0, 0.0D0,
-     +                  0.0D0, 0.0D0, 0.04D0, 0.0D0, 0.0D0, 0.0D0,
-     +                  0.0D0, 0.0D0, 0.0D0, 0.7D0, -1.08D0, 0.0D0,
-     +                  0.0D0, 0.0D0, 0.0D0, 0.0D0, 0.64D0, -1.26D0,
-     +                  0.54D0, 0.20D0, 0.0D0, 0.0D0, 0.0D0, 0.5D0,
-     +                  0.0D0, 0.0D0, 0.0D0, 0.0D0, 0.0D0, 0.0D0,
-     +                  0.04D0, 0.0D0, 0.0D0, 0.0D0, 0.0D0, 0.0D0,
-     +                  0.0D0, 0.04D0, -0.9D0, 0.18D0, 0.0D0, 0.0D0,
-     +                  0.0D0, 0.0D0, 0.04D0, -0.9D0, 0.18D0, 0.7D0,
-     +                  -0.18D0, 0.2D0, 0.16D0/
+      DATA              LEN/10/
+      DATA              PARAM/-2.0D0, 1.0D0, 0.0D0, 0.0D0, 1.0D0,
+     +                  -1.0D0, 0.2D0, 0.3D0, 0.4D0, 0.5D0,
+     +                  0.0D0, 1.0D0, 0.3D0, 0.4D0, 1.0D0,
+     +                  1.0D0, 0.2D0, -1.0D0, 1.0D0, 0.5D0/
       DATA              SSIZE2/0.0D0, 0.0D0, 0.0D0, 0.0D0, 0.0D0, 0.0D0,
-     +                  0.0D0, 0.0D0, 0.0D0, 0.0D0, 0.0D0, 0.0D0, 0.0D0,
-     +                  0.0D0, 1.17D0, 1.17D0, 1.17D0, 1.17D0, 1.17D0,
+     +                  0.0D0, 0.0D0, 0.0D0, 0.0D0, 1.17D0, 1.17D0, 
      +                  1.17D0, 1.17D0, 1.17D0, 1.17D0, 1.17D0, 1.17D0,
-     +                  1.17D0, 1.17D0, 1.17D0/
+     +                  1.17D0, 1.17D0/
 *     .. Executable Statements ..
 *
-      DO 60 KI = 1, 4
+      DO 60 KI = 1, 7
          INCX = INCXS(KI)
          INCY = INCYS(KI)
-         MX = ABS(INCX)
-         MY = ABS(INCY)
 *
-         DO 40 KN = 1, 4
+         DO 40 KN = 1, 5
             N = NS(KN)
             KSIZE = MIN(2,KN)
-            LENX = LENS(KN,MX)
-            LENY = LENS(KN,MY)
 *
             IF (ICASE.EQ.4) THEN
 *              .. DROTTEST ..
-               DO 20 I = 1, 7
-                  SX(I) = DX1(I)
-                  SY(I) = DY1(I)
-                  STX(I) = DT9X(I,KN,KI)
-                  STY(I) = DT9Y(I,KN,KI)
+               DO 20 I = 1, 10
+                  SX(I) = DX(I)
+                  SY(I) = DY(I)
+                  STX(I) = DX(I)
+                  STY(I) = DY(I)
    20          CONTINUE
                CALL DROTTEST(N,SX,INCX,SY,INCY,SC,SS)
-               CALL STEST(LENX,SX,STX,SSIZE2(1,KSIZE),SFAC)
-               CALL STEST(LENY,SY,STY,SSIZE2(1,KSIZE),SFAC)
-            ELSE
+               CALL DROT(N,STX,INCX,STY,INCY,SC,SS)
+               CALL STEST(LEN,SX,STX,SSIZE2(1,KSIZE),SFAC)
+               CALL STEST(LEN,SY,STY,SSIZE2(1,KSIZE),SFAC)
+            ELSE IF (ICASE.EQ.11) THEN
+*              .. DROTMTEST ..
+               DO 90 I = 1, 10
+                  SX(I) = DX(I)
+                  SY(I) = DY(I)
+                  STX(I) = DX(I)
+                  STY(I) = DY(I)
+   90          CONTINUE
+               DO 70 I = 1, 4
+                  DO 80 K = 1, 5
+                     DPARAM(K) = PARAM(K,I)
+   80             CONTINUE
+                  CALL DROTMTEST(N,SX,INCX,SY,INCY,DPARAM)
+                  CALL DROTM(N,STX,INCX,STY,INCY,DPARAM)
+                  CALL STEST(LEN,SX,STX,SSIZE2(1,KSIZE),SFAC)
+                  CALL STEST(LEN,SY,STY,SSIZE2(1,KSIZE),SFAC)
+   70          CONTINUE      
+               ELSE
                WRITE (NOUT,*) ' Shouldn''t be here in CHECK3'
                STOP
             END IF
    40    CONTINUE
    60 CONTINUE
-*
-      MWPC(1) = 1
-      DO 80 I = 2, 11
-         MWPC(I) = 0
-   80 CONTINUE
-      MWPS(1) = 0.0
-      DO 100 I = 2, 6
-         MWPS(I) = 1.0
-  100 CONTINUE
-      DO 120 I = 7, 11
-         MWPS(I) = -1.0
-  120 CONTINUE
-      MWPINX(1) = 1
-      MWPINX(2) = 1
-      MWPINX(3) = 1
-      MWPINX(4) = -1
-      MWPINX(5) = 1
-      MWPINX(6) = -1
-      MWPINX(7) = 1
-      MWPINX(8) = 1
-      MWPINX(9) = -1
-      MWPINX(10) = 1
-      MWPINX(11) = -1
-      MWPINY(1) = 1
-      MWPINY(2) = 1
-      MWPINY(3) = -1
-      MWPINY(4) = -1
-      MWPINY(5) = 2
-      MWPINY(6) = 1
-      MWPINY(7) = 1
-      MWPINY(8) = -1
-      MWPINY(9) = -1
-      MWPINY(10) = 2
-      MWPINY(11) = 1
-      DO 140 I = 1, 11
-         MWPN(I) = 5
-  140 CONTINUE
-      MWPN(5) = 3
-      MWPN(10) = 3
-      DO 160 I = 1, 5
-         MWPX(I) = I
-         MWPY(I) = I
-         MWPTX(1,I) = I
-         MWPTY(1,I) = I
-         MWPTX(2,I) = I
-         MWPTY(2,I) = -I
-         MWPTX(3,I) = 6 - I
-         MWPTY(3,I) = I - 6
-         MWPTX(4,I) = I
-         MWPTY(4,I) = -I
-         MWPTX(6,I) = 6 - I
-         MWPTY(6,I) = I - 6
-         MWPTX(7,I) = -I
-         MWPTY(7,I) = I
-         MWPTX(8,I) = I - 6
-         MWPTY(8,I) = 6 - I
-         MWPTX(9,I) = -I
-         MWPTY(9,I) = I
-         MWPTX(11,I) = I - 6
-         MWPTY(11,I) = 6 - I
-  160 CONTINUE
-      MWPTX(5,1) = 1
-      MWPTX(5,2) = 3
-      MWPTX(5,3) = 5
-      MWPTX(5,4) = 4
-      MWPTX(5,5) = 5
-      MWPTY(5,1) = -1
-      MWPTY(5,2) = 2
-      MWPTY(5,3) = -2
-      MWPTY(5,4) = 4
-      MWPTY(5,5) = -3
-      MWPTX(10,1) = -1
-      MWPTX(10,2) = -3
-      MWPTX(10,3) = -5
-      MWPTX(10,4) = 4
-      MWPTX(10,5) = 5
-      MWPTY(10,1) = 1
-      MWPTY(10,2) = 2
-      MWPTY(10,3) = 2
-      MWPTY(10,4) = 4
-      MWPTY(10,5) = 3
-      DO 200 I = 1, 11
-         INCX = MWPINX(I)
-         INCY = MWPINY(I)
-         DO 180 K = 1, 5
-            COPYX(K) = MWPX(K)
-            COPYY(K) = MWPY(K)
-            MWPSTX(K) = MWPTX(I,K)
-            MWPSTY(K) = MWPTY(I,K)
-  180    CONTINUE
-         CALL DROTTEST(MWPN(I),COPYX,INCX,COPYY,INCY,MWPC(I),MWPS(I))
-         CALL STEST(5,COPYX,MWPSTX,MWPSTX,SFAC)
-         CALL STEST(5,COPYY,MWPSTY,MWPSTY,SFAC)
-  200 CONTINUE
       RETURN
       END
       SUBROUTINE STEST(LEN,SCOMP,STRUE,SSIZE,SFAC)
@@ -726,3 +609,144 @@
      +       /1X)
 99997 FORMAT (1X,I4,I3,3I5,2I36,I12)
       END
+      SUBROUTINE DROT(N,DX,INCX,DY,INCY,C,S)
+*     .. Scalar Arguments ..
+      DOUBLE PRECISION C,S
+      INTEGER INCX,INCY,N
+*     ..
+*     .. Array Arguments ..
+      DOUBLE PRECISION DX(*),DY(*)
+*     ..
+*     applies a plane rotation.
+*     jack dongarra, linpack, 3/11/78.
+*     modified 12/3/93, array(1) declarations changed to array(*)
+*
+*     .. Local Scalars ..
+      DOUBLE PRECISION DTEMP
+      INTEGER I,IX,IY
+*     ..
+      IF (N.LE.0) RETURN
+      IF (INCX.EQ.1 .AND. INCY.EQ.1) GO TO 20
+      IX = 1
+      IY = 1
+      IF (INCX.LT.0) IX = (-N+1)*INCX + 1
+      IF (INCY.LT.0) IY = (-N+1)*INCY + 1
+      DO 10 I = 1,N
+          DTEMP = C*DX(IX) + S*DY(IY)
+          DY(IY) = C*DY(IY) - S*DX(IX)
+          DX(IX) = DTEMP
+          IX = IX + INCX
+          IY = IY + INCY
+   10 CONTINUE
+      RETURN
+   20 DO 30 I = 1,N
+          DTEMP = C*DX(I) + S*DY(I)
+          DY(I) = C*DY(I) - S*DX(I)
+          DX(I) = DTEMP
+   30 CONTINUE
+      RETURN
+      END
+      SUBROUTINE drotm(N,DX,INCX,DY,INCY,DPARAM)
+*
+*  -- Reference BLAS level1 routine (version 3.8.0) --
+*  -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
+*  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
+*     November 2017
+*
+*     .. Scalar Arguments ..
+      INTEGER INCX,INCY,N
+*     ..
+*     .. Array Arguments ..
+      DOUBLE PRECISION DPARAM(5),DX(*),DY(*)
+*     ..
+*
+*  =====================================================================
+*
+*     .. Local Scalars ..
+      DOUBLE PRECISION DFLAG,DH11,DH12,DH21,DH22,TWO,W,Z,ZERO
+      INTEGER I,KX,KY,NSTEPS
+*     ..
+*     .. Data statements ..
+      DATA zero,two/0.d0,2.d0/
+*     ..
+*
+      dflag = dparam(1)
+      IF (n.LE.0 .OR. (dflag+two.EQ.zero)) RETURN
+      IF (incx.EQ.incy.AND.incx.GT.0) THEN
+*
+         nsteps = n*incx
+         IF (dflag.LT.zero) THEN
+            dh11 = dparam(2)
+            dh12 = dparam(4)
+            dh21 = dparam(3)
+            dh22 = dparam(5)
+            DO i = 1,nsteps,incx
+               w = dx(i)
+               z = dy(i)
+               dx(i) = w*dh11 + z*dh12
+               dy(i) = w*dh21 + z*dh22
+            END DO
+         ELSE IF (dflag.EQ.zero) THEN
+            dh12 = dparam(4)
+            dh21 = dparam(3)
+            DO i = 1,nsteps,incx
+               w = dx(i)
+               z = dy(i)
+               dx(i) = w + z*dh12
+               dy(i) = w*dh21 + z
+            END DO
+         ELSE
+            dh11 = dparam(2)
+            dh22 = dparam(5)
+            DO i = 1,nsteps,incx
+               w = dx(i)
+               z = dy(i)
+               dx(i) = w*dh11 + z
+               dy(i) = -w + dh22*z
+            END DO
+         END IF
+      ELSE
+         kx = 1
+         ky = 1
+         IF (incx.LT.0) kx = 1 + (1-n)*incx
+         IF (incy.LT.0) ky = 1 + (1-n)*incy
+*
+         IF (dflag.LT.zero) THEN
+            dh11 = dparam(2)
+            dh12 = dparam(4)
+            dh21 = dparam(3)
+            dh22 = dparam(5)
+            DO i = 1,n
+               w = dx(kx)
+               z = dy(ky)
+               dx(kx) = w*dh11 + z*dh12
+               dy(ky) = w*dh21 + z*dh22
+               kx = kx + incx
+               ky = ky + incy
+            END DO
+         ELSE IF (dflag.EQ.zero) THEN
+            dh12 = dparam(4)
+            dh21 = dparam(3)
+            DO i = 1,n
+               w = dx(kx)
+               z = dy(ky)
+               dx(kx) = w + z*dh12
+               dy(ky) = w*dh21 + z
+               kx = kx + incx
+               ky = ky + incy
+            END DO
+         ELSE
+             dh11 = dparam(2)
+             dh22 = dparam(5)
+             DO i = 1,n
+                w = dx(kx)
+                z = dy(ky)
+                dx(kx) = w*dh11 + z
+                dy(ky) = -w + dh22*z
+                kx = kx + incx
+                ky = ky + incy
+            END DO
+         END IF
+      END IF
+      RETURN
+      END
diff --git a/ctest/c_sblas1.c b/ctest/c_sblas1.c
index 1f301a693..1a433b287 100644
--- a/ctest/c_sblas1.c
+++ b/ctest/c_sblas1.c
@@ -55,6 +55,13 @@ void F77_srotg( float *a, float *b, float *c, float *s)
    return;
 }
 
+void F77_srotm(blasint *N, float *X, blasint *incX, float *Y, blasint *incY,
+               float *param)
+{
+    cblas_srotm(*N, X, *incX, Y, *incY, param);
+    return;
+}
+
 void F77_srot( blasint *N, float *X, blasint *incX, float *Y,
               blasint *incY, const float  *c, const float  *s)
 {
diff --git a/ctest/c_sblat1.f b/ctest/c_sblat1.f
index 89902f12d..66a5def89 100644
--- a/ctest/c_sblat1.f
+++ b/ctest/c_sblat1.f
@@ -19,7 +19,7 @@
       DATA             SFAC/9.765625E-4/
 *     .. Executable Statements ..
       WRITE (NOUT,99999)
-      DO 20 IC = 1, 10
+      DO 20 IC = 1, 11
          ICASE = IC
          CALL HEADER
 *
@@ -40,7 +40,7 @@
          ELSE IF (ICASE.EQ.1 .OR. ICASE.EQ.2 .OR. ICASE.EQ.5 .OR.
      +            ICASE.EQ.6) THEN
             CALL CHECK2(SFAC)
-         ELSE IF (ICASE.EQ.4) THEN
+         ELSE IF (ICASE.EQ.4 .OR. ICASE.EQ.11) THEN
             CALL CHECK3(SFAC)
          END IF
 *        -- Print
@@ -59,7 +59,7 @@
       INTEGER          ICASE, INCX, INCY, MODE, N
       LOGICAL          PASS
 *     .. Local Arrays ..
-      CHARACTER*15      L(10)
+      CHARACTER*15      L(11)
 *     .. Common blocks ..
       COMMON           /COMBLA/ICASE, N, INCX, INCY, MODE, PASS
 *     .. Data statements ..
@@ -73,6 +73,7 @@
       DATA             L(8)/'CBLAS_SASUM '/
       DATA             L(9)/'CBLAS_SSCAL '/
       DATA             L(10)/'CBLAS_ISAMAX'/
+      DATA             L(11)/'CBLAS_SROTM'/
 *     .. Executable Statements ..
       WRITE (NOUT,99999) ICASE, L(ICASE)
       RETURN
@@ -396,203 +397,92 @@
 *     .. Scalar Arguments ..
       REAL              SFAC
 *     .. Scalars in Common ..
-      INTEGER           ICASE, INCX, INCY, MODE, N
+      INTEGER           ICASE, INCX, INCY, N
       LOGICAL           PASS
 *     .. Local Scalars ..
       REAL              SC, SS
-      INTEGER           I, K, KI, KN, KSIZE, LENX, LENY, MX, MY
+      INTEGER           I, K, KI, KN, KSIZE, LEN
 *     .. Local Arrays ..
-      REAL              COPYX(5), COPYY(5), DT9X(7,4,4), DT9Y(7,4,4),
-     +                  DX1(7), DY1(7), MWPC(11), MWPS(11), MWPSTX(5),
-     +                  MWPSTY(5), MWPTX(11,5), MWPTY(11,5), MWPX(5),
-     +                  MWPY(5), SSIZE2(14,2), STX(7), STY(7), SX(7),
-     +                  SY(7)
-      INTEGER           INCXS(4), INCYS(4), LENS(4,2), MWPINX(11),
-     +                  MWPINY(11), MWPN(11), NS(4)
+      REAL              DX(19), DY(19),
+     +                  SSIZE2(19,2), STX(19), STY(19), SX(19), SY(19),
+     +                  PARAM(5, 4), SPARAM(5) 
+      INTEGER           INCXS(7), INCYS(7), NS(7)
 *     .. External Subroutines ..
-      EXTERNAL          SROTTEST, STEST
+      EXTERNAL          SROTMTEST, SROTM
 *     .. Intrinsic Functions ..
-      INTRINSIC         ABS, MIN
+      INTRINSIC         MIN
 *     .. Common blocks ..
       COMMON            /COMBLA/ICASE, N, INCX, INCY, MODE, PASS
 *     .. Data statements ..
-      DATA              INCXS/1, 2, -2, -1/
-      DATA              INCYS/1, -2, 1, -2/
-      DATA              LENS/1, 1, 2, 4, 1, 1, 3, 7/
-      DATA              NS/0, 1, 2, 4/
-      DATA              DX1/0.6E0, 0.1E0, -0.5E0, 0.8E0, 0.9E0, -0.3E0,
-     +                  -0.4E0/
-      DATA              DY1/0.5E0, -0.9E0, 0.3E0, 0.7E0, -0.6E0, 0.2E0,
+      DATA              INCXS/1, 1, 2, 2, -2, -1, -2/
+      DATA              INCYS/1, 2, 2, -2, 1, -2, -2/
+      DATA              NS/0, 1, 2, 4, 5, 8, 9/
+      DATA              DX/0.6E0, 0.1E0, -0.5E0, 0.8E0, 0.9E0, -0.3E0,
+     +                  -0.4E0, 0.5E0, -0.9E0, 0.3E0, 0.7E0, -0.6E0,
+     +                  0.2E0, 0.8E0, -0.46E0, 0.78E0, -0.46E0, -0.22E0,
+     +                  1.06E0/
+      DATA              DY/0.5E0, -0.9E0, 0.3E0, 0.7E0, -0.6E0, 0.2E0,
+     +                  0.6E0, 0.1E0, -0.5E0, 0.8E0, 0.9E0, -0.3E0,
+     +                  0.96E0, 0.1E0, -0.76E0, 0.8E0, 0.90E0, 0.66E0,
      +                  0.8E0/
       DATA              SC, SS/0.8E0, 0.6E0/
-      DATA              DT9X/0.6E0, 0.0E0, 0.0E0, 0.0E0, 0.0E0, 0.0E0,
-     +                  0.0E0, 0.78E0, 0.0E0, 0.0E0, 0.0E0, 0.0E0,
-     +                  0.0E0, 0.0E0, 0.78E0, -0.46E0, 0.0E0, 0.0E0,
-     +                  0.0E0, 0.0E0, 0.0E0, 0.78E0, -0.46E0, -0.22E0,
-     +                  1.06E0, 0.0E0, 0.0E0, 0.0E0, 0.6E0, 0.0E0,
-     +                  0.0E0, 0.0E0, 0.0E0, 0.0E0, 0.0E0, 0.78E0,
-     +                  0.0E0, 0.0E0, 0.0E0, 0.0E0, 0.0E0, 0.0E0,
-     +                  0.66E0, 0.1E0, -0.1E0, 0.0E0, 0.0E0, 0.0E0,
-     +                  0.0E0, 0.96E0, 0.1E0, -0.76E0, 0.8E0, 0.90E0,
-     +                  -0.3E0, -0.02E0, 0.6E0, 0.0E0, 0.0E0, 0.0E0,
-     +                  0.0E0, 0.0E0, 0.0E0, 0.78E0, 0.0E0, 0.0E0,
-     +                  0.0E0, 0.0E0, 0.0E0, 0.0E0, -0.06E0, 0.1E0,
-     +                  -0.1E0, 0.0E0, 0.0E0, 0.0E0, 0.0E0, 0.90E0,
-     +                  0.1E0, -0.22E0, 0.8E0, 0.18E0, -0.3E0, -0.02E0,
-     +                  0.6E0, 0.0E0, 0.0E0, 0.0E0, 0.0E0, 0.0E0, 0.0E0,
-     +                  0.78E0, 0.0E0, 0.0E0, 0.0E0, 0.0E0, 0.0E0,
-     +                  0.0E0, 0.78E0, 0.26E0, 0.0E0, 0.0E0, 0.0E0,
-     +                  0.0E0, 0.0E0, 0.78E0, 0.26E0, -0.76E0, 1.12E0,
-     +                  0.0E0, 0.0E0, 0.0E0/
-      DATA              DT9Y/0.5E0, 0.0E0, 0.0E0, 0.0E0, 0.0E0, 0.0E0,
-     +                  0.0E0, 0.04E0, 0.0E0, 0.0E0, 0.0E0, 0.0E0,
-     +                  0.0E0, 0.0E0, 0.04E0, -0.78E0, 0.0E0, 0.0E0,
-     +                  0.0E0, 0.0E0, 0.0E0, 0.04E0, -0.78E0, 0.54E0,
-     +                  0.08E0, 0.0E0, 0.0E0, 0.0E0, 0.5E0, 0.0E0,
-     +                  0.0E0, 0.0E0, 0.0E0, 0.0E0, 0.0E0, 0.04E0,
-     +                  0.0E0, 0.0E0, 0.0E0, 0.0E0, 0.0E0, 0.0E0, 0.7E0,
-     +                  -0.9E0, -0.12E0, 0.0E0, 0.0E0, 0.0E0, 0.0E0,
-     +                  0.64E0, -0.9E0, -0.30E0, 0.7E0, -0.18E0, 0.2E0,
-     +                  0.28E0, 0.5E0, 0.0E0, 0.0E0, 0.0E0, 0.0E0,
-     +                  0.0E0, 0.0E0, 0.04E0, 0.0E0, 0.0E0, 0.0E0,
-     +                  0.0E0, 0.0E0, 0.0E0, 0.7E0, -1.08E0, 0.0E0,
-     +                  0.0E0, 0.0E0, 0.0E0, 0.0E0, 0.64E0, -1.26E0,
-     +                  0.54E0, 0.20E0, 0.0E0, 0.0E0, 0.0E0, 0.5E0,
-     +                  0.0E0, 0.0E0, 0.0E0, 0.0E0, 0.0E0, 0.0E0,
-     +                  0.04E0, 0.0E0, 0.0E0, 0.0E0, 0.0E0, 0.0E0,
-     +                  0.0E0, 0.04E0, -0.9E0, 0.18E0, 0.0E0, 0.0E0,
-     +                  0.0E0, 0.0E0, 0.04E0, -0.9E0, 0.18E0, 0.7E0,
-     +                  -0.18E0, 0.2E0, 0.16E0/
+      DATA              PARAM/-2.0E0, 1.0E0, 0.0E0, 0.0E0, 1.0E0,
+     +                  -1.0E0, 0.2E0, 0.3E0, 0.4E0, 0.5E0,
+     +                  0.0E0, 1.0E0, 0.3E0, 0.4E0, 1.0E0,
+     +                  1.0E0, 0.2E0, -1.0E0, 1.0E0, 0.5E0/
+      DATA              LEN/19/
       DATA              SSIZE2/0.0E0, 0.0E0, 0.0E0, 0.0E0, 0.0E0, 0.0E0,
      +                  0.0E0, 0.0E0, 0.0E0, 0.0E0, 0.0E0, 0.0E0, 0.0E0,
-     +                  0.0E0, 1.17E0, 1.17E0, 1.17E0, 1.17E0, 1.17E0,
+     +                  0.0E0, 0.0E0, 0.0E0, 0.0E0, 0.0E0, 0.0E0,
      +                  1.17E0, 1.17E0, 1.17E0, 1.17E0, 1.17E0, 1.17E0,
-     +                  1.17E0, 1.17E0, 1.17E0/
+     +                  1.17E0, 1.17E0, 1.17E0, 1.17E0, 1.17E0, 1.17E0,
+     +                  1.17E0, 1.17E0, 1.17E0, 1.17E0, 1.17E0, 1.17E0,
+     +                  1.17E0/
 *     .. Executable Statements ..
 *
-      DO 60 KI = 1, 4
+      DO 60 KI = 1, 7
          INCX = INCXS(KI)
          INCY = INCYS(KI)
-         MX = ABS(INCX)
-         MY = ABS(INCY)
 *
-         DO 40 KN = 1, 4
+         DO 40 KN = 1, 7
             N = NS(KN)
             KSIZE = MIN(2,KN)
-            LENX = LENS(KN,MX)
-            LENY = LENS(KN,MY)
 *
             IF (ICASE.EQ.4) THEN
 *              .. SROTTEST ..
-               DO 20 I = 1, 7
-                  SX(I) = DX1(I)
-                  SY(I) = DY1(I)
-                  STX(I) = DT9X(I,KN,KI)
-                  STY(I) = DT9Y(I,KN,KI)
+               DO 20 I = 1, 19
+                  SX(I) = DX(I)
+                  SY(I) = DY(I)
+                  STX(I) = DX(I)
+                  STY(I) = DY(I)
    20          CONTINUE
                CALL SROTTEST(N,SX,INCX,SY,INCY,SC,SS)
-               CALL STEST(LENX,SX,STX,SSIZE2(1,KSIZE),SFAC)
-               CALL STEST(LENY,SY,STY,SSIZE2(1,KSIZE),SFAC)
+               CALL SROT(N,STX,INCX,STY,INCY,SC,SS)
+               CALL STEST(LEN,SX,STX,SSIZE2(1,KSIZE),SFAC)
+               CALL STEST(LEN,SY,STY,SSIZE2(1,KSIZE),SFAC)
+            ELSE IF (ICASE.EQ.11) THEN
+*              .. SROTMTEST ..
+               DO 90 I = 1, 19
+                  SX(I) = DX(I)
+                  SY(I) = DY(I)
+                  STX(I) = DX(I)
+                  STY(I) = DY(I)
+   90          CONTINUE
+               DO 70 I = 1, 4
+                  DO 80 K = 1, 5
+                     SPARAM(K) = PARAM(K,I)
+   80             CONTINUE
+                  CALL SROTMTEST(N,SX,INCX,SY,INCY,SPARAM)
+                  CALL SROTM(N,STX,INCX,STY,INCY,SPARAM)
+                  CALL STEST(LEN,SX,STX,SSIZE2(1,KSIZE),SFAC)
+                  CALL STEST(LEN,SY,STY,SSIZE2(1,KSIZE),SFAC)
+   70          CONTINUE
             ELSE
                WRITE (NOUT,*) ' Shouldn''t be here in CHECK3'
                STOP
             END IF
    40    CONTINUE
    60 CONTINUE
-*
-      MWPC(1) = 1
-      DO 80 I = 2, 11
-         MWPC(I) = 0
-   80 CONTINUE
-      MWPS(1) = 0
-      DO 100 I = 2, 6
-         MWPS(I) = 1
-  100 CONTINUE
-      DO 120 I = 7, 11
-         MWPS(I) = -1
-  120 CONTINUE
-      MWPINX(1) = 1
-      MWPINX(2) = 1
-      MWPINX(3) = 1
-      MWPINX(4) = -1
-      MWPINX(5) = 1
-      MWPINX(6) = -1
-      MWPINX(7) = 1
-      MWPINX(8) = 1
-      MWPINX(9) = -1
-      MWPINX(10) = 1
-      MWPINX(11) = -1
-      MWPINY(1) = 1
-      MWPINY(2) = 1
-      MWPINY(3) = -1
-      MWPINY(4) = -1
-      MWPINY(5) = 2
-      MWPINY(6) = 1
-      MWPINY(7) = 1
-      MWPINY(8) = -1
-      MWPINY(9) = -1
-      MWPINY(10) = 2
-      MWPINY(11) = 1
-      DO 140 I = 1, 11
-         MWPN(I) = 5
-  140 CONTINUE
-      MWPN(5) = 3
-      MWPN(10) = 3
-      DO 160 I = 1, 5
-         MWPX(I) = I
-         MWPY(I) = I
-         MWPTX(1,I) = I
-         MWPTY(1,I) = I
-         MWPTX(2,I) = I
-         MWPTY(2,I) = -I
-         MWPTX(3,I) = 6 - I
-         MWPTY(3,I) = I - 6
-         MWPTX(4,I) = I
-         MWPTY(4,I) = -I
-         MWPTX(6,I) = 6 - I
-         MWPTY(6,I) = I - 6
-         MWPTX(7,I) = -I
-         MWPTY(7,I) = I
-         MWPTX(8,I) = I - 6
-         MWPTY(8,I) = 6 - I
-         MWPTX(9,I) = -I
-         MWPTY(9,I) = I
-         MWPTX(11,I) = I - 6
-         MWPTY(11,I) = 6 - I
-  160 CONTINUE
-      MWPTX(5,1) = 1
-      MWPTX(5,2) = 3
-      MWPTX(5,3) = 5
-      MWPTX(5,4) = 4
-      MWPTX(5,5) = 5
-      MWPTY(5,1) = -1
-      MWPTY(5,2) = 2
-      MWPTY(5,3) = -2
-      MWPTY(5,4) = 4
-      MWPTY(5,5) = -3
-      MWPTX(10,1) = -1
-      MWPTX(10,2) = -3
-      MWPTX(10,3) = -5
-      MWPTX(10,4) = 4
-      MWPTX(10,5) = 5
-      MWPTY(10,1) = 1
-      MWPTY(10,2) = 2
-      MWPTY(10,3) = 2
-      MWPTY(10,4) = 4
-      MWPTY(10,5) = 3
-      DO 200 I = 1, 11
-         INCX = MWPINX(I)
-         INCY = MWPINY(I)
-         DO 180 K = 1, 5
-            COPYX(K) = MWPX(K)
-            COPYY(K) = MWPY(K)
-            MWPSTX(K) = MWPTX(I,K)
-            MWPSTY(K) = MWPTY(I,K)
-  180    CONTINUE
-         CALL SROTTEST(MWPN(I),COPYX,INCX,COPYY,INCY,MWPC(I),MWPS(I))
-         CALL STEST(5,COPYX,MWPSTX,MWPSTX,SFAC)
-         CALL STEST(5,COPYY,MWPSTY,MWPSTY,SFAC)
-  200 CONTINUE
       RETURN
       END
       SUBROUTINE STEST(LEN,SCOMP,STRUE,SSIZE,SFAC)
@@ -726,3 +616,147 @@
      +       /1X)
 99997 FORMAT (1X,I4,I3,3I5,2I36,I12)
       END
+      SUBROUTINE SROT(N,SX,INCX,SY,INCY,C,S)
+*
+*   --Reference BLAS level1 routine (version 3.8.0) --
+*   --Reference BLAS is a software package provided by Univ. of Tennessee,    --
+*   --Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
+*     November 2017
+* 
+*     .. Scalar Arguments ..
+      REAL C,S
+      INTEGER INCX,INCY,N
+*     ..
+*     .. Array Arguments ..
+      REAL SX(*),SY(*)
+*     ..
+*     .. Local Scalars ..
+      REAL STEMP
+      INTEGER I,IX,IY
+*     ..
+      IF (n.LE.0) RETURN
+      IF (incx.EQ.1 .AND. incy.EQ.1) THEN
+         DO i = 1,n
+            stemp = c*sx(i) + s*sy(i)
+            sy(i) = c*sy(i) - s*sx(i)
+            sx(i) = stemp
+         END DO
+      ELSE
+         ix = 1
+         iy = 1
+         IF (incx.LT.0) ix = (-n+1)*incx + 1
+         IF (incy.LT.0) iy = (-n+1)*incy + 1
+         DO i = 1,n
+            stemp = c*sx(ix) + s*sy(iy)
+            sy(iy) = c*sy(iy) - s*sx(ix)
+            sx(ix) = stemp
+            ix = ix + incx
+            iy = iy + incy
+         END DO
+      END IF
+      RETURN
+      END
+      SUBROUTINE srotm(N,SX,INCX,SY,INCY,SPARAM)
+* 
+*   --Reference BLAS level1 routine (version 3.8.0) --
+*   --Reference BLAS is a software package provided by Univ. of Tennessee,    --
+*   --Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
+*     November 2017
+* 
+*     .. Scalar Arguments ..
+      INTEGER INCX,INCY,N
+*     ..
+*     .. Array Arguments ..
+      REAL SPARAM(5),SX(*),SY(*)
+*     ..
+* 
+*   ====================================================================
+* 
+*     .. Local Scalars ..
+      REAL SFLAG,SH11,SH12,SH21,SH22,TWO,W,Z,ZERO
+      INTEGER I,KX,KY,NSTEPS
+*     ..
+*     .. Data statements ..
+      DATA zero,two/0.e0,2.e0/
+*     ..
+* 
+      sflag = sparam(1)
+      IF (n.LE.0 .OR. (sflag+two.EQ.zero)) RETURN
+      IF (incx.EQ.incy.AND.incx.GT.0) THEN
+* 
+         nsteps = n*incx
+         IF (sflag.LT.zero) THEN
+            sh11 = sparam(2)
+            sh12 = sparam(4)
+            sh21 = sparam(3)
+            sh22 = sparam(5)
+            DO i = 1,nsteps,incx
+               w = sx(i)
+               z = sy(i)
+               sx(i) = w*sh11 + z*sh12
+               sy(i) = w*sh21 + z*sh22
+            END DO
+         ELSE IF (sflag.EQ.zero) THEN
+            sh12 = sparam(4)
+            sh21 = sparam(3)
+            DO i = 1,nsteps,incx
+               w = sx(i)
+               z = sy(i)
+               sx(i) = w + z*sh12
+               sy(i) = w*sh21 + z
+            END DO
+         ELSE
+            sh11 = sparam(2)
+            sh22 = sparam(5)
+            DO i = 1,nsteps,incx
+               w = sx(i)
+               z = sy(i)
+               sx(i) = w*sh11 + z
+               sy(i) = -w + sh22*z
+            END DO
+         END IF
+      ELSE
+         kx = 1
+         ky = 1
+         IF (incx.LT.0) kx = 1 + (1-n)*incx
+         IF (incy.LT.0) ky = 1 + (1-n)*incy
+* 
+         IF (sflag.LT.zero) THEN
+            sh11 = sparam(2)
+            sh12 = sparam(4)
+            sh21 = sparam(3)
+            sh22 = sparam(5)
+            DO i = 1,n
+               w = sx(kx)
+               z = sy(ky)
+               sx(kx) = w*sh11 + z*sh12
+               sy(ky) = w*sh21 + z*sh22
+               kx = kx + incx
+               ky = ky + incy
+            END DO
+         ELSE IF (sflag.EQ.zero) THEN
+            sh12 = sparam(4)
+            sh21 = sparam(3)
+            DO i = 1,n
+               w = sx(kx)
+               z = sy(ky)
+               sx(kx) = w + z*sh12
+               sy(ky) = w*sh21 + z
+               kx = kx + incx
+               ky = ky + incy
+            END DO
+         ELSE
+             sh11 = sparam(2)
+             sh22 = sparam(5)
+             DO i = 1,n
+                w = sx(kx)
+                z = sy(ky)
+                sx(kx) = w*sh11 + z
+                sy(ky) = -w + sh22*z
+                kx = kx + incx
+                ky = ky + incy
+            END DO
+         END IF
+      END IF
+      RETURN
+      END
\ No newline at end of file
diff --git a/driver/level2/gemv_thread.c b/driver/level2/gemv_thread.c
index d57740314..0d8c6b005 100644
--- a/driver/level2/gemv_thread.c
+++ b/driver/level2/gemv_thread.c
@@ -72,9 +72,9 @@
        defined __BORLANDC__ )
 #  define thread_local __declspec(thread) 
 /* note that ICC (linux) and Clang are covered by __GNUC__ */
-# elif defined __GNUC__ || \
+# elif (defined __GNUC__ || \
        defined __SUNPRO_C || \
-       defined __xlC__
+       defined __xlC__) && !defined(__APPLE__)
 #  define thread_local __thread
 # else
 # define UNSAFE
diff --git a/driver/level3/Makefile b/driver/level3/Makefile
index e320092e3..09a62d9bf 100644
--- a/driver/level3/Makefile
+++ b/driver/level3/Makefile
@@ -19,6 +19,10 @@ ifeq ($(ARCH), MIPS)
 USE_GEMM3M = 1
 endif
 
+ifeq ($(BUILD_HALF),1)
+SHBLASOBJS       += shgemm_nn.$(SUFFIX) shgemm_nt.$(SUFFIX) shgemm_tn.$(SUFFIX) shgemm_tt.$(SUFFIX)
+endif
+
 SBLASOBJS	+= \
 	sgemm_nn.$(SUFFIX) sgemm_nt.$(SUFFIX) sgemm_tn.$(SUFFIX) sgemm_tt.$(SUFFIX) \
 	strmm_LNUU.$(SUFFIX) strmm_LNUN.$(SUFFIX) strmm_LNLU.$(SUFFIX) strmm_LNLN.$(SUFFIX) \
@@ -203,7 +207,9 @@ COMMONOBJS  += gemm_thread_m.$(SUFFIX) gemm_thread_n.$(SUFFIX) gemm_thread_mn.$(
 COMMONOBJS  += syrk_thread.$(SUFFIX)
 
 ifndef USE_SIMPLE_THREADED_LEVEL3
-
+ifeq ($(BUILD_HALF),1)
+SHBLASOBJS    += shgemm_thread_nn.$(SUFFIX) shgemm_thread_nt.$(SUFFIX) shgemm_thread_tn.$(SUFFIX) shgemm_thread_tt.$(SUFFIX)
+endif
 SBLASOBJS    += sgemm_thread_nn.$(SUFFIX) sgemm_thread_nt.$(SUFFIX) sgemm_thread_tn.$(SUFFIX) sgemm_thread_tt.$(SUFFIX)
 DBLASOBJS    += dgemm_thread_nn.$(SUFFIX) dgemm_thread_nt.$(SUFFIX) dgemm_thread_tn.$(SUFFIX) dgemm_thread_tt.$(SUFFIX)
 QBLASOBJS    += qgemm_thread_nn.$(SUFFIX) qgemm_thread_nt.$(SUFFIX) qgemm_thread_tn.$(SUFFIX) qgemm_thread_tt.$(SUFFIX)
@@ -283,6 +289,18 @@ endif
 
 all ::
 
+shgemm_nn.$(SUFFIX) : gemm.c level3.c ../../param.h
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DHALF -UDOUBLE -UCOMPLEX -DNN $< -o $(@F)
+
+shgemm_nt.$(SUFFIX) : gemm.c level3.c ../../param.h
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DHALF -UDOUBLE -UCOMPLEX -DNT $< -o $(@F)
+
+shgemm_tn.$(SUFFIX) : gemm.c level3.c ../../param.h
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DHALF -UDOUBLE -UCOMPLEX -DTN $< -o $(@F)
+
+shgemm_tt.$(SUFFIX) : gemm.c level3.c ../../param.h
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DHALF -UDOUBLE -UCOMPLEX -DTT $< -o $(@F)
+
 sgemm_nn.$(SUFFIX) : gemm.c level3.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -UDOUBLE -UCOMPLEX -DNN $< -o $(@F)
 
@@ -478,6 +496,17 @@ gemm_thread_variable.$(SUFFIX) : gemm_thread_variable.c ../../common.h
 beta_thread.$(SUFFIX) : beta_thread.c ../../common.h
 	$(CC) -c $(CFLAGS) $< -o $(@F)
 
+shgemm_thread_nn.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DHALF -UDOUBLE -UCOMPLEX -DNN $< -o $(@F)
+
+shgemm_thread_nt.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DHALF -UDOUBLE -UCOMPLEX -DNT $< -o $(@F)
+
+shgemm_thread_tn.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DHALF -UDOUBLE -UCOMPLEX -DTN $< -o $(@F)
+
+shgemm_thread_tt.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
+	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DHALF -UDOUBLE -UCOMPLEX -DTT $< -o $(@F)
 
 sgemm_thread_nn.$(SUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(CFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -UCOMPLEX -DNN $< -o $(@F)
@@ -2652,6 +2681,18 @@ xtrsm_RCLU.$(SUFFIX) : trsm_R.c
 xtrsm_RCLN.$(SUFFIX) : trsm_R.c
 	$(CC) -c $(CFLAGS) -DCOMPLEX -DXDOUBLE -DTRANSA -UUPPER -UUNIT -DCONJ $< -o $(@F)
 
+shgemm_nn.$(PSUFFIX) : gemm.c level3.c ../../param.h
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DHALF -UDOUBLE -UCOMPLEX -DNN $< -o $(@F)
+
+shgemm_nt.$(PSUFFIX) : gemm.c level3.c ../../param.h
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DHALF -UDOUBLE -UCOMPLEX -DNT $< -o $(@F)
+
+shgemm_tn.$(PSUFFIX) : gemm.c level3.c ../../param.h
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DHALF -UDOUBLE -UCOMPLEX -DTN $< -o $(@F)
+
+shgemm_tt.$(PSUFFIX) : gemm.c level3.c ../../param.h
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DHALF -UDOUBLE -UCOMPLEX -DTT $< -o $(@F)
+
 sgemm_nn.$(PSUFFIX) : gemm.c level3.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -UDOUBLE -UCOMPLEX -DNN $< -o $(@F)
 
@@ -2848,6 +2889,18 @@ beta_thread.$(PSUFFIX) : beta_thread.c ../../common.h
 	$(CC) -c $(PFLAGS) $< -o $(@F)
 
 
+shgemm_thread_nn.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DHALF -UDOUBLE -UCOMPLEX -DNN $< -o $(@F)
+
+shgemm_thread_nt.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DHALF -UDOUBLE -UCOMPLEX -DNT $< -o $(@F)
+
+shgemm_thread_tn.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DHALF -UDOUBLE -UCOMPLEX -DTN $< -o $(@F)
+
+shgemm_thread_tt.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
+	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -DHALF -UDOUBLE -UCOMPLEX -DTT $< -o $(@F)
+
 sgemm_thread_nn.$(PSUFFIX) : gemm.c level3_thread.c ../../param.h
 	$(CC) $(PFLAGS) $(BLOCKS) -c -DTHREADED_LEVEL3 -UDOUBLE -UCOMPLEX -DNN $< -o $(@F)
 
diff --git a/driver/level3/level3.c b/driver/level3/level3.c
index 9aa67286f..c6bbb9ca9 100644
--- a/driver/level3/level3.c
+++ b/driver/level3/level3.c
@@ -62,18 +62,18 @@
 #ifndef ICOPY_OPERATION
 #if defined(NN) || defined(NT) || defined(NC) || defined(NR) || \
     defined(RN) || defined(RT) || defined(RC) || defined(RR)
-#define ICOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_ITCOPY(M, N, (FLOAT *)(A) + ((Y) + (X) * (LDA)) * COMPSIZE, LDA, BUFFER);
+#define ICOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_ITCOPY(M, N, (IFLOAT *)(A) + ((Y) + (X) * (LDA)) * COMPSIZE, LDA, BUFFER);
 #else
-#define ICOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_INCOPY(M, N, (FLOAT *)(A) + ((X) + (Y) * (LDA)) * COMPSIZE, LDA, BUFFER);
+#define ICOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_INCOPY(M, N, (IFLOAT *)(A) + ((X) + (Y) * (LDA)) * COMPSIZE, LDA, BUFFER);
 #endif
 #endif
 
 #ifndef OCOPY_OPERATION
 #if defined(NN) || defined(TN) || defined(CN) || defined(RN) || \
     defined(NR) || defined(TR) || defined(CR) || defined(RR)
-#define OCOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_ONCOPY(M, N, (FLOAT *)(A) + ((X) + (Y) * (LDA)) * COMPSIZE, LDA, BUFFER);
+#define OCOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_ONCOPY(M, N, (IFLOAT *)(A) + ((X) + (Y) * (LDA)) * COMPSIZE, LDA, BUFFER);
 #else
-#define OCOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_OTCOPY(M, N, (FLOAT *)(A) + ((Y) + (X) * (LDA)) * COMPSIZE, LDA, BUFFER);
+#define OCOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_OTCOPY(M, N, (IFLOAT *)(A) + ((Y) + (X) * (LDA)) * COMPSIZE, LDA, BUFFER);
 #endif
 #endif
 
@@ -173,7 +173,8 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
 		  XFLOAT *sa, XFLOAT *sb, BLASLONG dummy){
   BLASLONG k, lda, ldb, ldc;
   FLOAT *alpha, *beta;
-  FLOAT *a, *b, *c;
+  IFLOAT *a, *b;
+  FLOAT *c;
   BLASLONG m_from, m_to, n_from, n_to;
 
   BLASLONG ls, is, js;
@@ -198,8 +199,8 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
 
   k = K;
 
-  a = (FLOAT *)A;
-  b = (FLOAT *)B;
+  a = (IFLOAT *)A;
+  b = (IFLOAT *)B;
   c = (FLOAT *)C;
 
   lda = LDA;
diff --git a/driver/level3/level3_thread.c b/driver/level3/level3_thread.c
index ca0085e71..5a8d497d2 100644
--- a/driver/level3/level3_thread.c
+++ b/driver/level3/level3_thread.c
@@ -117,18 +117,18 @@ typedef struct {
 #ifndef ICOPY_OPERATION
 #if defined(NN) || defined(NT) || defined(NC) || defined(NR) || \
   defined(RN) || defined(RT) || defined(RC) || defined(RR)
-#define ICOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_ITCOPY(M, N, (FLOAT *)(A) + ((Y) + (X) * (LDA)) * COMPSIZE, LDA, BUFFER);
+#define ICOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_ITCOPY(M, N, (IFLOAT *)(A) + ((Y) + (X) * (LDA)) * COMPSIZE, LDA, BUFFER);
 #else
-#define ICOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_INCOPY(M, N, (FLOAT *)(A) + ((X) + (Y) * (LDA)) * COMPSIZE, LDA, BUFFER);
+#define ICOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_INCOPY(M, N, (IFLOAT *)(A) + ((X) + (Y) * (LDA)) * COMPSIZE, LDA, BUFFER);
 #endif
 #endif
 
 #ifndef OCOPY_OPERATION
 #if defined(NN) || defined(TN) || defined(CN) || defined(RN) || \
   defined(NR) || defined(TR) || defined(CR) || defined(RR)
-#define OCOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_ONCOPY(M, N, (FLOAT *)(A) + ((X) + (Y) * (LDA)) * COMPSIZE, LDA, BUFFER);
+#define OCOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_ONCOPY(M, N, (IFLOAT *)(A) + ((X) + (Y) * (LDA)) * COMPSIZE, LDA, BUFFER);
 #else
-#define OCOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_OTCOPY(M, N, (FLOAT *)(A) + ((Y) + (X) * (LDA)) * COMPSIZE, LDA, BUFFER);
+#define OCOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_OTCOPY(M, N, (IFLOAT *)(A) + ((Y) + (X) * (LDA)) * COMPSIZE, LDA, BUFFER);
 #endif
 #endif
 
@@ -219,15 +219,16 @@ typedef struct {
 #define STOP_RPCC(COUNTER)
 #endif
 
-static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLOAT *sb, BLASLONG mypos){
+static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, IFLOAT *sa, IFLOAT *sb, BLASLONG mypos){
 
-  FLOAT *buffer[DIVIDE_RATE];
+  IFLOAT *buffer[DIVIDE_RATE];
 
   BLASLONG k, lda, ldb, ldc;
   BLASLONG m_from, m_to, n_from, n_to;
 
   FLOAT *alpha, *beta;
-  FLOAT *a, *b, *c;
+  IFLOAT *a, *b;
+  FLOAT *c;
   job_t *job = (job_t *)args -> common;
 
   BLASLONG nthreads_m;
@@ -255,8 +256,8 @@ static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
 
   k = K;
 
-  a = (FLOAT *)A;
-  b = (FLOAT *)B;
+  a = (IFLOAT *)A;
+  b = (IFLOAT *)B;
   c = (FLOAT *)C;
 
   lda = LDA;
@@ -425,7 +426,7 @@ static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
           /* Apply kernel with local region of A and part of other region of B */
 	  START_RPCC();
 	  KERNEL_OPERATION(min_i, MIN(range_n[current + 1]  - js,  div_n), min_l, alpha,
-			   sa, (FLOAT *)job[current].working[mypos][CACHE_LINE_SIZE * bufferside],
+			   sa, (IFLOAT *)job[current].working[mypos][CACHE_LINE_SIZE * bufferside],
 			   c, ldc, m_from, js);
           STOP_RPCC(kernel);
 
@@ -469,7 +470,7 @@ static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
           /* Apply kernel with local region of A and part of region of B */
 	  START_RPCC();
 	  KERNEL_OPERATION(min_i, MIN(range_n[current + 1] - js, div_n), min_l, alpha,
-			   sa, (FLOAT *)job[current].working[mypos][CACHE_LINE_SIZE * bufferside],
+			   sa, (IFLOAT *)job[current].working[mypos][CACHE_LINE_SIZE * bufferside],
 			   c, ldc, is, js);
           STOP_RPCC(kernel);
           
@@ -532,7 +533,7 @@ static int round_up(int remainder, int width, int multiple)
 
 
 static int gemm_driver(blas_arg_t *args, BLASLONG *range_m, BLASLONG
-		       *range_n, FLOAT *sa, FLOAT *sb,
+		       *range_n, IFLOAT *sa, IFLOAT *sb,
                        BLASLONG nthreads_m, BLASLONG nthreads_n) {
 
 #ifndef USE_OPENMP
@@ -728,7 +729,7 @@ EnterCriticalSection((PCRITICAL_SECTION)&level3_lock);
   return 0;
 }
 
-int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLOAT *sb, BLASLONG mypos){
+int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, IFLOAT *sa, IFLOAT *sb, BLASLONG mypos){
 
   BLASLONG m = args -> m;
   BLASLONG n = args -> n;
diff --git a/driver/others/blas_server.c b/driver/others/blas_server.c
index 6f4e20610..04b614a6e 100644
--- a/driver/others/blas_server.c
+++ b/driver/others/blas_server.c
@@ -72,6 +72,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "common.h"
 #if defined(OS_LINUX) || defined(OS_NETBSD) || defined(OS_DARWIN) || defined(OS_ANDROID) || defined(OS_SUNOS) || defined(OS_FREEBSD) || defined(OS_OPENBSD) || defined(OS_DRAGONFLY) || defined(OS_HAIKU)
 #include <dlfcn.h>
+#include <errno.h>
 #include <signal.h>
 #include <sys/resource.h>
 #include <sys/time.h>
@@ -140,6 +141,16 @@ typedef struct {
 
 } thread_status_t;
 
+#if (__STDC_VERSION__ >= 201112L)
+#define	atomic_load_queue(p)		__atomic_load_n(p, __ATOMIC_RELAXED)
+#define	atomic_store_queue(p, v)	__atomic_store_n(p, v, __ATOMIC_RELAXED)
+#else
+#define	atomic_load_queue(p)		(blas_queue_t*)(*(volatile blas_queue_t**)(p))
+#define	atomic_store_queue(p, v)	(*(volatile blas_queue_t* volatile*)(p) = (v))
+#endif
+
+
+
 static thread_status_t thread_status[MAX_CPU_NUMBER] __attribute__((aligned(ATTRIBUTE_SIZE)));
 
 #ifndef THREAD_TIMEOUT
@@ -261,7 +272,7 @@ static void legacy_exec(void *func, int mode, blas_arg_t *args, void *sb){
       }
 }
 
-#if defined(OS_LINUX) && !defined(NO_AFFINITY)
+#if defined(OS_LINUX) && !defined(NO_AFFINITY) 
 int gotoblas_set_affinity(int);
 int gotoblas_set_affinity2(int);
 int get_node(void);
@@ -269,6 +280,25 @@ int get_node(void);
 
 static int increased_threads = 0;
 
+#ifdef OS_LINUX
+extern int openblas_get_num_threads(void);  
+
+int openblas_setaffinity(int thread_idx, size_t cpusetsize, cpu_set_t* cpu_set) {
+  const int active_threads = openblas_get_num_threads();
+
+  if (thread_idx < 0 || thread_idx >= active_threads) {
+    errno = EINVAL;
+    return -1;
+  }
+
+  pthread_t thread = (thread_idx == active_threads - 1)
+      ? pthread_self()
+      : blas_threads[thread_idx];
+
+  return pthread_setaffinity_np(thread, cpusetsize, cpu_set);
+}
+#endif
+
 static void* blas_thread_server(void *arg){
 
   /* Thread identifier */
@@ -312,20 +342,19 @@ blas_queue_t *tscq;
 
       last_tick = (unsigned int)rpcc();
 
-	pthread_mutex_lock  (&thread_status[cpu].lock);
-        tscq=thread_status[cpu].queue;
-	pthread_mutex_unlock  (&thread_status[cpu].lock);
+      tscq = atomic_load_queue(&thread_status[cpu].queue);
 
 	while(!tscq) {
 	YIELDING;
 
 	if ((unsigned int)rpcc() - last_tick > thread_timeout) {
 
-	  pthread_mutex_lock  (&thread_status[cpu].lock);
 
-	  if (!thread_status[cpu].queue) {
+	  if (!atomic_load_queue(&thread_status[cpu].queue)) {
+	    pthread_mutex_lock  (&thread_status[cpu].lock);
 	    thread_status[cpu].status = THREAD_STATUS_SLEEP;
-	    while (thread_status[cpu].status == THREAD_STATUS_SLEEP) {
+	    while (thread_status[cpu].status == THREAD_STATUS_SLEEP && 
+			    !atomic_load_queue(&thread_status[cpu].queue)) {
 
 #ifdef MONITOR
 	      main_status[cpu] = MAIN_SLEEPING;
@@ -333,19 +362,18 @@ blas_queue_t *tscq;
 
 	      pthread_cond_wait(&thread_status[cpu].wakeup, &thread_status[cpu].lock);
 	    }
+	    pthread_mutex_unlock(&thread_status[cpu].lock);
 	  }
 
-	  pthread_mutex_unlock(&thread_status[cpu].lock);
-
 	  last_tick = (unsigned int)rpcc();
 	}
-	pthread_mutex_lock  (&thread_status[cpu].lock);
-        tscq=thread_status[cpu].queue;
-	pthread_mutex_unlock  (&thread_status[cpu].lock);
+
+	tscq = atomic_load_queue(&thread_status[cpu].queue);
 
       }
 
-    queue = thread_status[cpu].queue;
+      queue = atomic_load_queue(&thread_status[cpu].queue);
+      MB;
 
     if ((long)queue == -1) break;
 
@@ -360,9 +388,7 @@ blas_queue_t *tscq;
     if (queue) {
       int (*routine)(blas_arg_t *, void *, void *, void *, void *, BLASLONG) = queue -> routine;
 
-      pthread_mutex_lock  (&thread_status[cpu].lock);
-      thread_status[cpu].queue = (blas_queue_t *)1;
-      pthread_mutex_unlock  (&thread_status[cpu].lock);
+      atomic_store_queue(&thread_status[cpu].queue, (blas_queue_t *)1);
 
       sa = queue -> sa;
       sb = queue -> sb;
@@ -442,13 +468,9 @@ blas_queue_t *tscq;
 
       // arm: make sure all results are written out _before_
       // thread is marked as done and other threads use them
-      WMB;
+      MB;
+      atomic_store_queue(&thread_status[cpu].queue, (blas_queue_t *)0);
 
-      pthread_mutex_lock  (&thread_status[cpu].lock);
-      thread_status[cpu].queue = (blas_queue_t * volatile) ((long)thread_status[cpu].queue & 0);  /* Need a trick */
-      pthread_mutex_unlock  (&thread_status[cpu].lock);
-      
-      WMB;
 
     }
 
@@ -566,7 +588,7 @@ int blas_thread_init(void){
 
     for(i = 0; i < blas_num_threads - 1; i++){
 
-      thread_status[i].queue  = (blas_queue_t *)NULL;
+      atomic_store_queue(&thread_status[i].queue, (blas_queue_t *)0);
       thread_status[i].status = THREAD_STATUS_WAKEUP;
 
       pthread_mutex_init(&thread_status[i].lock, NULL);
@@ -582,7 +604,7 @@ int blas_thread_init(void){
       if(ret!=0){
 	struct rlimit rlim;
         const char *msg = strerror(ret);
-        fprintf(STDERR, "OpenBLAS blas_thread_init: pthread_create failed for thread %ld of %ld: %s\n", i+1,blas_num_threads,msg);
+        fprintf(STDERR, "OpenBLAS blas_thread_init: pthread_create failed for thread %ld of %d: %s\n", i+1,blas_num_threads,msg);
 #ifdef RLIMIT_NPROC
         if(0 == getrlimit(RLIMIT_NPROC, &rlim)) {
           fprintf(STDERR, "OpenBLAS blas_thread_init: RLIMIT_NPROC "
@@ -655,7 +677,8 @@ int exec_blas_async(BLASLONG pos, blas_queue_t *queue){
       if (queue -> mode & BLAS_NODE) {
 
 	do {
-	  while((thread_status[i].node != node || thread_status[i].queue) && (i < blas_num_threads - 1)) i ++;
+      
+	  while((thread_status[i].node != node || atomic_load_queue(&thread_status[i].queue)) && (i < blas_num_threads - 1)) i ++;
 
 	  if (i < blas_num_threads - 1) break;
 
@@ -669,36 +692,26 @@ int exec_blas_async(BLASLONG pos, blas_queue_t *queue){
 	} while (1);
 
       } else {
-	pthread_mutex_lock (&thread_status[i].lock);
-	tsiq = thread_status[i].queue;
-	pthread_mutex_unlock (&thread_status[i].lock);      
+	tsiq = atomic_load_queue(&thread_status[i].queue);
 	while(tsiq) {
 	  i ++;
 	  if (i >= blas_num_threads - 1) i = 0;
-	  pthread_mutex_lock (&thread_status[i].lock);
-	  tsiq = thread_status[i].queue;
-	  pthread_mutex_unlock (&thread_status[i].lock);
+	  tsiq = atomic_load_queue(&thread_status[i].queue);
 	}
       }
 #else
-      pthread_mutex_lock  (&thread_status[i].lock);
-      tsiq=thread_status[i].queue ;
-      pthread_mutex_unlock  (&thread_status[i].lock);
+      tsiq = atomic_load_queue(&thread_status[i].queue);
       while(tsiq) {
 	i ++;
 	if (i >= blas_num_threads - 1) i = 0;
-        pthread_mutex_lock  (&thread_status[i].lock);
-        tsiq=thread_status[i].queue ;
-        pthread_mutex_unlock  (&thread_status[i].lock);
+        tsiq = atomic_load_queue(&thread_status[i].queue);
       }
 #endif
 
       queue -> assigned = i;
-      WMB;
-      pthread_mutex_lock  (&thread_status[i].lock);
-      thread_status[i].queue = queue;
-      pthread_mutex_unlock  (&thread_status[i].lock);
-      WMB;
+      MB;
+
+      atomic_store_queue(&thread_status[i].queue, queue);
 
       queue = queue -> next;
       pos ++;
@@ -718,9 +731,7 @@ int exec_blas_async(BLASLONG pos, blas_queue_t *queue){
 
       pos = current -> assigned;
 
-      pthread_mutex_lock  (&thread_status[pos].lock);
-      tspq=thread_status[pos].queue;
-      pthread_mutex_unlock  (&thread_status[pos].lock);
+      tspq = atomic_load_queue(&thread_status[pos].queue);
 
       if ((BLASULONG)tspq > 1) {
 	pthread_mutex_lock  (&thread_status[pos].lock);
@@ -752,24 +763,20 @@ int exec_blas_async_wait(BLASLONG num, blas_queue_t *queue){
 
     while ((num > 0) && queue) {
 
-      pthread_mutex_lock(&thread_status[queue->assigned].lock);
-      tsqq=thread_status[queue -> assigned].queue;
-      pthread_mutex_unlock(&thread_status[queue->assigned].lock);
+      tsqq = atomic_load_queue(&thread_status[queue->assigned].queue);
 
 
       while(tsqq) {
 	YIELDING;
-        pthread_mutex_lock(&thread_status[queue->assigned].lock);
-        tsqq=thread_status[queue -> assigned].queue;
-        pthread_mutex_unlock(&thread_status[queue->assigned].lock);
-
-
+        tsqq = atomic_load_queue(&thread_status[queue->assigned].queue);
       };
 
       queue = queue -> next;
       num --;
     }
 
+    MB;
+
 #ifdef SMP_DEBUG
   fprintf(STDERR, "Done.\n\n");
 #endif
@@ -880,7 +887,7 @@ void goto_set_num_threads(int num_threads) {
 
     for(i = blas_num_threads - 1; i < num_threads - 1; i++){
 
-      thread_status[i].queue  = (blas_queue_t *)NULL;
+      atomic_store_queue(&thread_status[i].queue, (blas_queue_t *)0);
       thread_status[i].status = THREAD_STATUS_WAKEUP;
 
       pthread_mutex_init(&thread_status[i].lock, NULL);
@@ -971,12 +978,11 @@ int BLASFUNC(blas_thread_shutdown)(void){
 
   for (i = 0; i < blas_num_threads - 1; i++) {
 
+
     pthread_mutex_lock (&thread_status[i].lock);
 
-    thread_status[i].queue = (blas_queue_t *)-1;
-
+    atomic_store_queue(&thread_status[i].queue, (blas_queue_t *)-1);
     thread_status[i].status = THREAD_STATUS_WAKEUP;
-
     pthread_cond_signal (&thread_status[i].wakeup);
 
     pthread_mutex_unlock(&thread_status[i].lock);
diff --git a/driver/others/dynamic.c b/driver/others/dynamic.c
index 2e87e186a..1bf0e4a6d 100644
--- a/driver/others/dynamic.c
+++ b/driver/others/dynamic.c
@@ -764,18 +764,53 @@ char *gotoblas_corename(void) {
   if (gotoblas == &gotoblas_NORTHWOOD)    return corename[ 3];
   if (gotoblas == &gotoblas_PRESCOTT)     return corename[ 4];
   if (gotoblas == &gotoblas_BANIAS)       return corename[ 5];
-  if (gotoblas == &gotoblas_ATOM)         return corename[ 6];
+  if (gotoblas == &gotoblas_ATOM)
+#ifdef DYNAMIC_OLDER
+           return corename[ 6];
+#else
+           return corename[10];
+#endif
   if (gotoblas == &gotoblas_CORE2)        return corename[ 7];
-  if (gotoblas == &gotoblas_PENRYN)       return corename[ 8];
-  if (gotoblas == &gotoblas_DUNNINGTON)   return corename[ 9];
+  if (gotoblas == &gotoblas_PENRYN)
+#ifdef DYNAMIC_OLDER
+           return corename[ 8];
+#else
+           return corename[7];
+#endif
+  if (gotoblas == &gotoblas_DUNNINGTON)
+#ifdef DYNAMIC_OLDER
+           return corename[ 9];
+#else
+           return corename[7];
+#endif
   if (gotoblas == &gotoblas_NEHALEM)      return corename[10];
   if (gotoblas == &gotoblas_ATHLON)       return corename[11];
-  if (gotoblas == &gotoblas_OPTERON_SSE3) return corename[12];
-  if (gotoblas == &gotoblas_OPTERON)      return corename[13];
+  if (gotoblas == &gotoblas_OPTERON_SSE3)
+#ifdef DYNAMIC_OLDER
+           return corename[12];
+#else
+           return corename[7];
+#endif
+  if (gotoblas == &gotoblas_OPTERON)
+#ifdef DYNAMIC_OLDER
+           return corename[13];
+#else
+           return corename[7];
+#endif
   if (gotoblas == &gotoblas_BARCELONA)    return corename[14];
-  if (gotoblas == &gotoblas_NANO)         return corename[15];
+  if (gotoblas == &gotoblas_NANO)
+#ifdef DYNAMIC_OLDER
+           return corename[15];
+#else
+           return corename[10];
+#endif
   if (gotoblas == &gotoblas_SANDYBRIDGE)  return corename[16];
-  if (gotoblas == &gotoblas_BOBCAT)       return corename[17];
+  if (gotoblas == &gotoblas_BOBCAT)
+#ifdef DYNAMIC_OLDER
+           return corename[17];
+#else
+           return corename[7];
+#endif
   if (gotoblas == &gotoblas_BULLDOZER)    return corename[18];
   if (gotoblas == &gotoblas_PILEDRIVER)   return corename[19];
   if (gotoblas == &gotoblas_HASWELL)      return corename[20];
@@ -787,6 +822,7 @@ char *gotoblas_corename(void) {
 }
 
 
+
 static gotoblas_t *force_coretype(char *coretype){
 
 	int i ;
diff --git a/driver/others/dynamic_power.c b/driver/others/dynamic_power.c
index 1dec5f4b3..8c831b998 100644
--- a/driver/others/dynamic_power.c
+++ b/driver/others/dynamic_power.c
@@ -3,7 +3,7 @@
 
 extern gotoblas_t gotoblas_POWER6;
 extern gotoblas_t gotoblas_POWER8;
-#if (!defined C_GCC) || (GCC_VERSION >= 60000)
+#if (!defined __GNUC__) || ( __GNUC__ >= 6)
 extern gotoblas_t gotoblas_POWER9;
 #endif
 
@@ -21,7 +21,7 @@ static char *corename[] = {
 char *gotoblas_corename(void) {
 	if (gotoblas == &gotoblas_POWER6)	return corename[1];
 	if (gotoblas == &gotoblas_POWER8)	return corename[2];
-#if (!defined C_GCC) || (GCC_VERSION >= 60000)
+#if (!defined __GNUC__) || ( __GNUC__ >= 6)
 	if (gotoblas == &gotoblas_POWER9)	return corename[3];
 #endif
 	return corename[0];
@@ -33,7 +33,7 @@ static gotoblas_t *get_coretype(void) {
 		return &gotoblas_POWER6;
 	if (__builtin_cpu_is("power8"))
 		return &gotoblas_POWER8;
-#if (!defined C_GCC) || (GCC_VERSION >= 60000)
+#if (!defined __GNUC__) || ( __GNUC__ >= 6)
 	if (__builtin_cpu_is("power9"))
 		return &gotoblas_POWER9;
 #endif
@@ -59,7 +59,7 @@ static gotoblas_t *force_coretype(char * coretype) {
 	{
 	case  1: return (&gotoblas_POWER6);
 	case  2: return (&gotoblas_POWER8);
-#if (!defined C_GCC) || (GCC_VERSION >= 60000)
+#if (!defined __GNUC__) || ( __GNUC__ >= 6)
 	case  3: return (&gotoblas_POWER9);
 #endif
 	default: return NULL;
diff --git a/driver/others/dynamic_zarch.c b/driver/others/dynamic_zarch.c
index 90d3051b1..403b34111 100644
--- a/driver/others/dynamic_zarch.c
+++ b/driver/others/dynamic_zarch.c
@@ -1,12 +1,58 @@
-
 #include "common.h"
+#include <stdbool.h>
 
+// Gate kernels for z13 and z14 on gcc version
+#if (__GNUC__ == 5 && __GNUC_MINOR__ >= 2) || __GNUC__ >= 6 ||           \
+    /* RHEL 7 since 7.3: */                                              \
+    (__GNUC__ == 4 && __GNUC_MINOR__ == 8 && __GNUC_PATCHLEVEL__ == 5 && \
+     __GNUC_RH_RELEASE__ >= 11)
+#define HAVE_Z13_SUPPORT
+#endif
+
+#if __GNUC__ >= 7
+#define HAVE_Z14_SUPPORT
+#endif
+
+// Guard the use of getauxval() on glibc version >= 2.16
+#ifdef __GLIBC__
+#include <features.h>
+#if __GLIBC_PREREQ(2, 16)
+#include <sys/auxv.h>
+#define HAVE_GETAUXVAL 1
+
+static unsigned long get_hwcap(void)
+{
+	unsigned long hwcap = getauxval(AT_HWCAP);
+	char *maskenv;
+
+	// honor requests for not using specific CPU features in LD_HWCAP_MASK
+	maskenv = getenv("LD_HWCAP_MASK");
+	if (maskenv)
+		hwcap &= strtoul(maskenv, NULL, 0);
+
+	return hwcap;
+	// note that a missing auxval is interpreted as no capabilities
+	// available, which is safe.
+}
+
+#else // __GLIBC_PREREQ(2, 16)
+#warn "Cannot detect SIMD support in Z13 or newer architectures since glibc is older than 2.16"
+
+static unsigned long get_hwcap(void) {
+	// treat missing support for getauxval() as no capabilities available,
+	// which is safe.
+	return 0;
+}
+#endif // __GLIBC_PREREQ(2, 16)
+#endif // __GLIBC
+
+extern gotoblas_t gotoblas_ZARCH_GENERIC;
+#ifdef HAVE_Z13_SUPPORT
 extern gotoblas_t gotoblas_Z13;
+#endif
+#ifdef HAVE_Z14_SUPPORT
 extern gotoblas_t gotoblas_Z14;
-//extern gotoblas_t gotoblas_Z15;
-//#if (!defined C_GCC) || (GCC_VERSION >= 60000)
-//extern gotoblas_t gotoblas_Z14;
-//#endif
+#endif
 
 #define NUM_CORETYPES 4
 
@@ -16,47 +62,50 @@ static char* corename[] = {
 	"unknown",
 	"Z13",
 	"Z14",
-//	"Z15",
 	"ZARCH_GENERIC",
 };
 
 char* gotoblas_corename(void) {
+#ifdef HAVE_Z13_SUPPORT
 	if (gotoblas == &gotoblas_Z13)	return corename[1];
+#endif
+#ifdef HAVE_Z14_SUPPORT
 	if (gotoblas == &gotoblas_Z14)	return corename[2];
-//	if (gotoblas == &gotoblas_Z15)	return corename[3];
-//#if (!defined C_GCC) || (GCC_VERSION >= 60000)
-//	if (gotoblas == &gotoblas_POWER9)	return corename[3];
-//#endif
-	return corename[0]; // try generic?
+#endif
+	if (gotoblas == &gotoblas_ZARCH_GENERIC) return corename[3];
+
+	return corename[0];
 }
 
-// __builtin_cpu_is is not supported by zarch
+/**
+ * Detect the fitting set of kernels by retrieving the CPU features supported by
+ * OS from the auxiliary value AT_HWCAP and choosing the set of kernels
+ * ("coretype") that exploits most of the features and can be compiled with the
+ * available gcc version.
+ * Note that we cannot use vector registers on a z13 or newer unless supported
+ * by the OS kernel (which needs to handle them properly during context switch).
+ */
 static gotoblas_t* get_coretype(void) {
-	FILE* infile;
-	char buffer[512], * p;
 
-	p = (char*)NULL;
-	infile = fopen("/proc/sysinfo", "r");
-	while (fgets(buffer, sizeof(buffer), infile)) {
-		if (!strncmp("Type", buffer, 4)) {
-			p = strchr(buffer, ':') + 2;
-#if 0
-			fprintf(stderr, "%s\n", p);
+	unsigned long hwcap __attribute__((unused)) = get_hwcap();
+
+	// z14 and z15 systems: exploit Vector Facility (SIMD) and
+	// Vector-Enhancements Facility 1 (float SIMD instructions), if present.
+#ifdef HAVE_Z14_SUPPORT
+	if ((hwcap & HWCAP_S390_VX) && (hwcap & HWCAP_S390_VXE))
+		return &gotoblas_Z14;
 #endif
-			break;
-		}
-	}
 
-	fclose(infile);
+	// z13: Vector Facility (SIMD for double)
+#ifdef HAVE_Z13_SUPPORT
+	if (hwcap & HWCAP_S390_VX)
+		return &gotoblas_Z13;
+#endif
 
-	if (strstr(p, "2964")) return &gotoblas_Z13;
-	if (strstr(p, "2965")) return &gotoblas_Z13;
-	if (strstr(p, "3906")) return &gotoblas_Z14;
-	if (strstr(p, "3907")) return &gotoblas_Z14;
-	if (strstr(p, "8561")) return &gotoblas_Z14;        // fallback z15 to z14
-	if (strstr(p, "8562")) return &gotoblas_Z14;        // fallback z15 to z14
-
-	return NULL; // should be ZARCH_GENERIC
+	// fallback in case of missing compiler support, systems before z13, or
+	// when the OS does not advertise support for the Vector Facility (e.g.,
+	// missing support in the OS kernel)
+	return &gotoblas_ZARCH_GENERIC;
 }
 
 static gotoblas_t* force_coretype(char* coretype) {
@@ -76,12 +125,13 @@ static gotoblas_t* force_coretype(char* coretype) {
 
 	switch (found)
 	{
+#ifdef HAVE_Z13_SUPPORT
 	case  1: return (&gotoblas_Z13);
+#endif
+#ifdef HAVE_Z14_SUPPORT
 	case  2: return (&gotoblas_Z14);
-//	case  3: return (&gotoblas_Z15);
-//#if (!defined C_GCC) || (GCC_VERSION >= 60000)
-//	case  3: return (&gotoblas_POWER9);
-//#endif
+#endif
+	case  3: return (&gotoblas_ZARCH_GENERIC);
 	default: return NULL;
 	}
 	snprintf(message, 128, "Core not found: %s\n", coretype);
@@ -109,9 +159,9 @@ void gotoblas_dynamic_init(void) {
 
 	if (gotoblas == NULL)
 	{
-		snprintf(coremsg, 128, "Falling back to Z14 core\n");
+		snprintf(coremsg, 128, "Failed to detect system, falling back to generic z support.\n");
 		openblas_warning(1, coremsg);
-		gotoblas = &gotoblas_Z14;
+		gotoblas = &gotoblas_ZARCH_GENERIC;
 	}
 
 	if (gotoblas && gotoblas->init) {
diff --git a/driver/others/memory.c b/driver/others/memory.c
index 62a5a0214..a5595aed4 100644
--- a/driver/others/memory.c
+++ b/driver/others/memory.c
@@ -87,6 +87,30 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #endif
 #endif
 
+/* Memory buffer must fit two matrix subblocks of maximal size */
+#define XSTR(x) STR(x)
+#define STR(x) #x
+#if BUFFER_SIZE < (SGEMM_DEFAULT_P * SGEMM_DEFAULT_Q * 4 * 2) || \
+    BUFFER_SIZE < (SGEMM_DEFAULT_P * SGEMM_DEFAULT_R * 4 * 2) || \
+    BUFFER_SIZE < (SGEMM_DEFAULT_R * SGEMM_DEFAULT_Q * 4 * 2)
+#warning BUFFER_SIZE is too small for P, Q, and R of SGEMM - large calculations may crash !
+#endif
+#if BUFFER_SIZE < (DGEMM_DEFAULT_P * DGEMM_DEFAULT_Q * 8 * 2) || \
+    BUFFER_SIZE < (DGEMM_DEFAULT_P * DGEMM_DEFAULT_R * 8 * 2) || \
+    BUFFER_SIZE < (DGEMM_DEFAULT_R * DGEMM_DEFAULT_Q * 8 * 2)
+#warning BUFFER_SIZE is too small for P, Q, and R of DGEMM - large calculations may crash !
+#endif
+#if BUFFER_SIZE < (CGEMM_DEFAULT_P * CGEMM_DEFAULT_Q * 8 * 2) || \
+    BUFFER_SIZE < (CGEMM_DEFAULT_P * CGEMM_DEFAULT_R * 8 * 2) || \
+    BUFFER_SIZE < (CGEMM_DEFAULT_R * CGEMM_DEFAULT_Q * 8 * 2)
+#warning BUFFER_SIZE is too small for P, Q, and R of CGEMM - large calculations may crash !
+#endif
+#if BUFFER_SIZE < (ZGEMM_DEFAULT_P * ZGEMM_DEFAULT_Q * 16 * 2) || \
+    BUFFER_SIZE < (ZGEMM_DEFAULT_P * ZGEMM_DEFAULT_R * 16 * 2) || \
+    BUFFER_SIZE < (ZGEMM_DEFAULT_R * ZGEMM_DEFAULT_Q * 16 * 2)
+#warning BUFFER_SIZE is too small for P, Q, and R of ZGEMM - large calculations may crash !
+#endif
+
 #if defined(COMPILE_TLS)
 
 #include <errno.h>
@@ -2046,7 +2070,7 @@ if (!release->address) return;
   if (munmap(release -> address, BUFFER_SIZE)) {
       int errsv=errno;
        perror("OpenBLAS : munmap failed:");
-       printf("error code=%d,\trelease->address=%lx\n",errsv,release->address);
+       printf("error code=%d,\trelease->address=%p\n",errsv,release->address);
   }
 }
 
@@ -2717,6 +2741,7 @@ void *blas_memory_alloc(int procpos){
   LOCK_COMMAND(&alloc_lock);
 #endif
   do {
+	  RMB;
 #if defined(USE_OPENMP)	  
     if (!memory[position].used) { 
       blas_lock(&memory[position].lock);
diff --git a/driver/others/parameter.c b/driver/others/parameter.c
index 8bf7da78b..b1f3befae 100644
--- a/driver/others/parameter.c
+++ b/driver/others/parameter.c
@@ -62,6 +62,11 @@ BLASLONG gemm_offset_b = DEFAULT_GEMM_OFFSET_B;
 BLASLONG gemm_offset_b = GEMM_OFFSET_B;
 #endif
 
+#if SHGEMM_P == shgemm_p
+BLASLONG shgemm_p = DEFAULT_GEMM_P;
+#else
+BLASLONG shgemm_p = SHGEMM_P;
+#endif
 #if SGEMM_P == sgemm_p
 BLASLONG sgemm_p = DEFAULT_GEMM_P;
 #else
@@ -83,6 +88,11 @@ BLASLONG zgemm_p = DEFAULT_GEMM_P;
 BLASLONG zgemm_p = ZGEMM_P;
 #endif
 
+#if SHGEMM_Q == shgemm_q
+BLASLONG shgemm_q = DEFAULT_GEMM_Q;
+#else
+BLASLONG shgemm_q = SHGEMM_Q;
+#endif
 #if SGEMM_Q == sgemm_q
 BLASLONG sgemm_q = DEFAULT_GEMM_Q;
 #else
@@ -104,6 +114,11 @@ BLASLONG zgemm_q = DEFAULT_GEMM_Q;
 BLASLONG zgemm_q = ZGEMM_Q;
 #endif
 
+#if SHGEMM_R == shgemm_r
+BLASLONG shgemm_r = DEFAULT_GEMM_R;
+#else
+BLASLONG shgemm_r = SHGEMM_R;
+#endif
 #if SGEMM_R == sgemm_r
 BLASLONG sgemm_r = DEFAULT_GEMM_R;
 #else
@@ -597,6 +612,7 @@ void blas_set_parameter(void){
 
   size = BITMASK(cpuid3, 16, 0xff);
 
+  shgemm_p = 192 * (size + 1);
   sgemm_p = 192 * (size + 1);
   dgemm_p =  96 * (size + 1);
   cgemm_p =  96 * (size + 1);
@@ -610,6 +626,7 @@ void blas_set_parameter(void){
   xgemm_p =  16 * (size + 1);
 #endif
 
+  shgemm_r = (((BUFFER_SIZE - ((SHGEMM_P * SHGEMM_Q *  4 + GEMM_OFFSET_A + GEMM_ALIGN) & ~GEMM_ALIGN)) / (SHGEMM_Q *  4)) - 15) & ~15;
   sgemm_r = (((BUFFER_SIZE - ((SGEMM_P * SGEMM_Q *  4 + GEMM_OFFSET_A + GEMM_ALIGN) & ~GEMM_ALIGN)) / (SGEMM_Q *  4)) - 15) & ~15;
   dgemm_r = (((BUFFER_SIZE - ((DGEMM_P * DGEMM_Q *  8 + GEMM_OFFSET_A + GEMM_ALIGN) & ~GEMM_ALIGN)) / (DGEMM_Q *  8)) - 15) & ~15;
   cgemm_r = (((BUFFER_SIZE - ((CGEMM_P * CGEMM_Q *  8 + GEMM_OFFSET_A + GEMM_ALIGN) & ~GEMM_ALIGN)) / (CGEMM_Q *  8)) - 15) & ~15;
diff --git a/exports/Makefile b/exports/Makefile
index d32e449df..01a313b35 100644
--- a/exports/Makefile
+++ b/exports/Makefile
@@ -30,6 +30,10 @@ ifndef BUILD_LAPACK_DEPRECATED
 BUILD_LAPACK_DEPRECATED = 0
 endif
 
+ifndef BUILD_HALF
+BUILD_HALF = 0
+endif
+
 ifeq ($(OSNAME), WINNT)
 ifeq ($(F_COMPILER), GFORTRAN)
 ifndef ONLY_CBLAS
@@ -126,7 +130,7 @@ endif
 dllinit.$(SUFFIX) : dllinit.c
 	$(CC) $(CFLAGS) -c -o $(@F) -s $<
 
-ifeq ($(OSNAME), $(filter $(OSNAME),Linux SunOS Android Haiku))
+ifeq ($(OSNAME), $(filter $(OSNAME),Linux SunOS Android Haiku FreeBSD DragonFly))
 
 so : ../$(LIBSONAME)
 
@@ -151,8 +155,12 @@ ifeq ($(F_COMPILER), INTEL)
 	-Wl,--whole-archive $< -Wl,--no-whole-archive \
 	-Wl,-soname,$(INTERNALNAME) $(EXTRALIB)
 	$(CC) $(CFLAGS) $(LDFLAGS) -w -o linktest linktest.c ../$(LIBSONAME) $(FEXTRALIB) && echo OK.
+else ifeq ($(F_COMPILER), FLANG)
+	$(FC) $(FFLAGS) $(LDFLAGS) -shared -o ../$(LIBSONAME) \
+	-Wl,--whole-archive $< -Wl,--no-whole-archive \
+	-Wl,-soname,$(INTERNALNAME) $(EXTRALIB)
+	$(CC) $(CFLAGS) $(LDFLAGS) -w -o linktest linktest.c ../$(LIBSONAME) $(FEXTRALIB) && echo OK.
 else
-
 ifneq ($(C_COMPILER), LSB)
 	$(CC) $(CFLAGS) $(LDFLAGS) -shared -o ../$(LIBSONAME) \
 	-Wl,--whole-archive $< -Wl,--no-whole-archive \
@@ -171,7 +179,7 @@ endif
 endif
 
 #http://stackoverflow.com/questions/7656425/makefile-ifeq-logical-or
-ifeq ($(OSNAME), $(filter $(OSNAME),FreeBSD OpenBSD NetBSD DragonFly))
+ifeq ($(OSNAME), $(filter $(OSNAME),OpenBSD NetBSD))
 
 so : ../$(LIBSONAME)
 
@@ -234,23 +242,23 @@ static : ../$(LIBNAME)
 	rm -f goto.$(SUFFIX)
 
 osx.def : gensymbol ../Makefile.system ../getarch.c
-	perl ./gensymbol osx $(ARCH) $(BU) $(EXPRECISION) $(NO_CBLAS)  $(NO_LAPACK) $(NO_LAPACKE) $(NEED2UNDERSCORES) $(ONLY_CBLAS) "$(SYMBOLPREFIX)" "$(SYMBOLSUFFIX)" $(BUILD_LAPACK_DEPRECATED) > $(@F)
+	perl ./gensymbol osx $(ARCH) $(BU) $(EXPRECISION) $(NO_CBLAS)  $(NO_LAPACK) $(NO_LAPACKE) $(NEED2UNDERSCORES) $(ONLY_CBLAS) "$(SYMBOLPREFIX)" "$(SYMBOLSUFFIX)" $(BUILD_LAPACK_DEPRECATED) $(BUILD_HALF) > $(@F)
 
 aix.def : gensymbol ../Makefile.system ../getarch.c
-	perl ./gensymbol aix $(ARCH) $(BU) $(EXPRECISION) $(NO_CBLAS)  $(NO_LAPACK) $(NO_LAPACKE) $(NEED2UNDERSCORES) $(ONLY_CBLAS) "$(SYMBOLPREFIX)" "$(SYMBOLSUFFIX)" $(BUILD_LAPACK_DEPRECATED) > $(@F)
+	perl ./gensymbol aix $(ARCH) $(BU) $(EXPRECISION) $(NO_CBLAS)  $(NO_LAPACK) $(NO_LAPACKE) $(NEED2UNDERSCORES) $(ONLY_CBLAS) "$(SYMBOLPREFIX)" "$(SYMBOLSUFFIX)" $(BUILD_LAPACK_DEPRECATED) $(BUILD_HALF) > $(@F)
 
 objcopy.def : gensymbol ../Makefile.system ../getarch.c
-	perl ./gensymbol objcopy $(ARCH) $(BU) $(EXPRECISION) $(NO_CBLAS)  $(NO_LAPACK) $(NO_LAPACKE) $(NEED2UNDERSCORES) $(ONLY_CBLAS) "$(SYMBOLPREFIX)" "$(SYMBOLSUFFIX)" $(BUILD_LAPACK_DEPRECATED) > $(@F)
+	perl ./gensymbol objcopy $(ARCH) $(BU) $(EXPRECISION) $(NO_CBLAS)  $(NO_LAPACK) $(NO_LAPACKE) $(NEED2UNDERSCORES) $(ONLY_CBLAS) "$(SYMBOLPREFIX)" "$(SYMBOLSUFFIX)" $(BUILD_LAPACK_DEPRECATED) $(BUILD_HALF) > $(@F)
 
 objconv.def : gensymbol ../Makefile.system ../getarch.c
-	perl ./gensymbol objconv $(ARCH) $(BU) $(EXPRECISION) $(NO_CBLAS)  $(NO_LAPACK) $(NO_LAPACKE) $(NEED2UNDERSCORES) $(ONLY_CBLAS) "$(SYMBOLPREFIX)" "$(SYMBOLSUFFIX)" $(BUILD_LAPACK_DEPRECATED) > $(@F)
+	perl ./gensymbol objconv $(ARCH) $(BU) $(EXPRECISION) $(NO_CBLAS)  $(NO_LAPACK) $(NO_LAPACKE) $(NEED2UNDERSCORES) $(ONLY_CBLAS) "$(SYMBOLPREFIX)" "$(SYMBOLSUFFIX)" $(BUILD_LAPACK_DEPRECATED) $(BUILD_HALF) > $(@F)
 
 test : linktest.c
 	$(CC) $(CFLAGS) $(LDFLAGS) -w -o linktest linktest.c ../$(LIBSONAME) -lm && echo OK.
 	rm -f linktest
 
 linktest.c : gensymbol ../Makefile.system ../getarch.c
-	perl ./gensymbol linktest  $(ARCH) $(BU) $(EXPRECISION) $(NO_CBLAS) $(NO_LAPACK) $(NO_LAPACKE) $(NEED2UNDERSCORES) $(ONLY_CBLAS) "$(SYMBOLPREFIX)" "$(SYMBOLSUFFIX)" $(BUILD_LAPACK_DEPRECATED) > linktest.c
+	perl ./gensymbol linktest  $(ARCH) $(BU) $(EXPRECISION) $(NO_CBLAS) $(NO_LAPACK) $(NO_LAPACKE) $(NEED2UNDERSCORES) $(ONLY_CBLAS) "$(SYMBOLPREFIX)" "$(SYMBOLSUFFIX)" $(BUILD_LAPACK_DEPRECATED) $(BUILD_HALF) > linktest.c
 
 clean ::
 	@rm -f *.def *.dylib __.SYMDEF* *.renamed
diff --git a/exports/gensymbol b/exports/gensymbol
index d2894e6c8..73b4be248 100644
--- a/exports/gensymbol
+++ b/exports/gensymbol
@@ -40,17 +40,13 @@
     ztbsv,ztpmv,ztpsv,ztrmm,ztrmv,ztrsm,ztrsv,
     xerbla,
     saxpby,daxpby,caxpby,zaxpby,
+    somatcopy, domatcopy, comatcopy, zomatcopy,
+    simatcopy, dimatcopy, cimatcopy, zimatcopy,
     sgeadd,dgeadd,cgeadd,zgeadd,
-    somatcopy,
-    simatcopy,
-    domatcopy,
-    dimatcopy,
-    comatcopy,
-    cimatcopy,
-    zomatcopy,
-    zimatcopy,
+    ssum, dsum, scsum, dzsum
 );
 
+@halfblasobjs = (shgemm);
 @cblasobjs = (
     cblas_caxpy, cblas_ccopy, cblas_cdotc, cblas_cdotu, cblas_cgbmv, cblas_cgemm, cblas_cgemv,
     cblas_cgerc, cblas_cgeru, cblas_chbmv, cblas_chemm, cblas_chemv, cblas_cher2, cblas_cher2k,
@@ -80,9 +76,16 @@
     cblas_saxpby,cblas_daxpby,cblas_caxpby,cblas_zaxpby,
     cblas_somatcopy, cblas_domatcopy, cblas_comatcopy, cblas_zomatcopy,
     cblas_simatcopy, cblas_dimatcopy, cblas_cimatcopy, cblas_zimatcopy,
-    cblas_sgeadd, cblas_dgeadd,cblas_cgeadd, cblas_zgeadd
+    cblas_sgeadd, cblas_dgeadd,cblas_cgeadd, cblas_zgeadd,
+    cblas_isamin, cblas_idamin, cblas_icamin, cblas_izamin,
+    cblas_ismin, cblas_idmin, cblas_icmin, cblas_izmin,
+    cblas_ismax, cblas_idmax, cblas_icmax, cblas_izmax,
+    cblas_ssum, cblas_dsum, cblas_scsum, cblas_dzsum,
+    cblas_xerbla
 );
 
+@halfcblasobjs = (cblas_shgemm);
+
 @exblasobjs = (
     qamax,qamin,qasum,qaxpy,qcabs1,qcopy,qdot,qgbmv,qgemm,
     qgemv,qger,qmax,qmin,
@@ -3454,6 +3457,10 @@ use File::Spec;
 use File::Basename;
 my $dirname = File::Spec->catfile(dirname(dirname(File::Spec->rel2abs(__FILE__))), "lapack-netlib");
 
+if ($ARGV[12] == 1) {
+	@blasobjs = (@blasobjs, @halfblasobjs);
+	@cblasobjs = (@cblasobjs, @halfcblasobjs);
+}
 if ($ARGV[8] == 1) {
     #ONLY_CBLAS=1
     @underscore_objs = (@misc_underscore_objs);
@@ -3494,9 +3501,12 @@ if ($ARGV[1] eq "x86")    { @underscore_objs = (@underscore_objs, @gemm3mobjs);
 if ($ARGV[1] eq "ia64")   { @underscore_objs = (@underscore_objs, @gemm3mobjs); };
 if ($ARGV[1] eq "MIPS")   { @underscore_objs = (@underscore_objs, @gemm3mobjs); };
 
-
 if ($ARGV[4] == 0) {
     @no_underscore_objs = (@cblasobjs, @misc_no_underscore_objs);
+    if ($ARGV[1] eq "x86_64") { @no_underscore_objs = (@no_underscore_objs, @cblasgemm3mobjs); };
+    if ($ARGV[1] eq "x86")    { @no_underscore_objs = (@no_underscore_objs, @cblasgemm3mobjs); };
+    if ($ARGV[1] eq "ia64")   { @no_underscore_objs = (@no_underscore_objs, @cblasgemm3mobjs); };
+    if ($ARGV[1] eq "MIPS")   { @no_underscore_objs = (@no_underscore_objs, @cblasgemm3mobjs); };
 }else{
     #NO_CBLAS=1
     @no_underscore_objs = (@misc_no_underscore_objs);
diff --git a/f_check b/f_check
index fac8fc707..17d863224 100644
--- a/f_check
+++ b/f_check
@@ -334,7 +334,8 @@ if ($link ne "") {
 	    && ($flags !~ /kernel32/)
 	    && ($flags !~ /advapi32/)
 	    && ($flags !~ /shell32/)
-	    && ($flags !~ /omp/)
+	    && ($flags !~ /omp/ || ($vendor !~ /PGI/ && $flags =~ /omp/))
+	    && ($flags !~ /[0-9]+/)
 		&& ($flags !~ /^\-l$/)
 	    ) {
 	    $linker_l .= $flags . " ";
diff --git a/getarch.c b/getarch.c
index 30ca290e3..c173d58b8 100644
--- a/getarch.c
+++ b/getarch.c
@@ -812,6 +812,34 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #else
 #endif
 
+#ifdef FORCE_MIPS1004K
+#define FORCE
+#define ARCHITECTURE    "MIPS"
+#define SUBARCHITECTURE "MIPS1004K"
+#define SUBDIRNAME      "mips"
+#define ARCHCONFIG   "-DMIPS1004K " \
+       "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=32 " \
+       "-DL2_SIZE=262144 -DL2_LINESIZE=32 " \
+       "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 -DL2_ASSOCIATIVE=8 "
+#define LIBNAME   "mips1004K"
+#define CORENAME  "MIPS1004K"
+#else
+#endif
+
+#ifdef FORCE_MIPS24K
+#define FORCE
+#define ARCHITECTURE    "MIPS"
+#define SUBARCHITECTURE "MIPS24K"
+#define SUBDIRNAME      "mips"
+#define ARCHCONFIG   "-DMIPS24K " \
+       "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=32 " \
+       "-DL2_SIZE=32768 -DL2_LINESIZE=32 " \
+       "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 -DL2_ASSOCIATIVE=8 "
+#define LIBNAME   "mips24K"
+#define CORENAME  "MIPS24K"
+#else
+#endif
+
 #ifdef FORCE_I6500
 #define FORCE
 #define ARCHITECTURE    "MIPS"
@@ -1038,7 +1066,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
        "-DL1_DATA_SIZE=65536 -DL1_DATA_LINESIZE=64 -DL1_DATA_ASSOCIATIVE=4 " \
        "-DL2_SIZE=1048576 -DL2_LINESIZE=64 -DL2_ASSOCIATIVE=16 " \
        "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 " \
-       "-DHAVE_VFPV4 -DHAVE_VFPV3 -DHAVE_VFP -DHAVE_NEON -DARMV8" \
+       "-DHAVE_VFPV4 -DHAVE_VFPV3 -DHAVE_VFP -DHAVE_NEON -DARMV8 " \
        "-march=armv8.2-a -mtune=cortex-a72"
 #define LIBNAME   "neoversen1"
 #define CORENAME  "NEOVERSEN1"
@@ -1334,10 +1362,12 @@ int main(int argc, char *argv[]){
 
 #if defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
 printf("__BYTE_ORDER__=__ORDER_BIG_ENDIAN__\n");
-#endif
-#if defined(__BIG_ENDIAN__) && __BIG_ENDIAN__ > 0
+#elif defined(__BIG_ENDIAN__) && __BIG_ENDIAN__ > 0
 printf("__BYTE_ORDER__=__ORDER_BIG_ENDIAN__\n");
 #endif
+#if defined(_CALL_ELF) && (_CALL_ELF == 2)
+printf("ELF_VERSION=2\n");
+#endif
 
 #ifdef MAKE_NB_JOBS
   #if MAKE_NB_JOBS > 0
diff --git a/getarch_2nd.c b/getarch_2nd.c
index cf9c578cb..a1d0ccac8 100644
--- a/getarch_2nd.c
+++ b/getarch_2nd.c
@@ -9,6 +9,8 @@
 int main(int argc, char **argv) {
 
   if ( (argc <= 1) || ((argc >= 2) && (*argv[1] == '0'))) {
+    printf("SHGEMM_UNROLL_M=%d\n", SHGEMM_DEFAULT_UNROLL_M);
+    printf("SHGEMM_UNROLL_N=%d\n", SHGEMM_DEFAULT_UNROLL_N);
     printf("SGEMM_UNROLL_M=%d\n", SGEMM_DEFAULT_UNROLL_M);
     printf("SGEMM_UNROLL_N=%d\n", SGEMM_DEFAULT_UNROLL_N);
     printf("DGEMM_UNROLL_M=%d\n", DGEMM_DEFAULT_UNROLL_M);
diff --git a/interface/CMakeLists.txt b/interface/CMakeLists.txt
index 5ea39f864..7a8fc6698 100644
--- a/interface/CMakeLists.txt
+++ b/interface/CMakeLists.txt
@@ -115,7 +115,7 @@ foreach (float_type ${FLOAT_TYPES})
     GenerateNamedObjects("syr2k.c" "HEMM" "her2k" ${CBLAS_FLAG} "" "" false ${float_type})
 
     if (USE_GEMM3M)
-      GenerateNamedObjects("gemm.c" "GEMM3M" "gemm3m" false "" "" false ${float_type})
+      GenerateNamedObjects("gemm.c" "GEMM3M" "gemm3m" ${CBLAS_FLAG} "" "" false ${float_type})
     endif()
   endif ()
   if (${float_type} STREQUAL "COMPLEX")
diff --git a/interface/Makefile b/interface/Makefile
index 3f0dcca28..44a9fdcf0 100644
--- a/interface/Makefile
+++ b/interface/Makefile
@@ -46,6 +46,9 @@ SBLAS3OBJS    = \
 		somatcopy.$(SUFFIX) simatcopy.$(SUFFIX)\
 		sgeadd.$(SUFFIX)
 
+ifeq ($(BUILD_HALF),1)
+SHBLAS3OBJS    = shgemm.$(SUFFIX)
+endif
 
 DBLAS1OBJS    = \
 		daxpy.$(SUFFIX) dswap.$(SUFFIX) \
@@ -277,6 +280,10 @@ CSBLAS3OBJS   = \
 	cblas_ssyrk.$(SUFFIX) cblas_ssyr2k.$(SUFFIX) cblas_somatcopy.$(SUFFIX)  cblas_simatcopy.$(SUFFIX)\
 	cblas_sgeadd.$(SUFFIX)
 
+ifeq ($(BUILD_HALF),1)
+CSHBLAS3OBJS = cblas_shgemm.$(SUFFIX)
+endif
+
 CDBLAS1OBJS   = \
 	cblas_idamax.$(SUFFIX) cblas_idamin.$(SUFFIX) cblas_dasum.$(SUFFIX) cblas_daxpy.$(SUFFIX) \
 	cblas_dcopy.$(SUFFIX) cblas_ddot.$(SUFFIX) \
@@ -367,6 +374,7 @@ override CFLAGS += -I.
 SBLAS1OBJS   += $(CSBLAS1OBJS)
 SBLAS2OBJS   += $(CSBLAS2OBJS)
 SBLAS3OBJS   += $(CSBLAS3OBJS)
+SHBLAS3OBJS  += $(CSHBLAS3OBJS)
 DBLAS1OBJS   += $(CDBLAS1OBJS)
 DBLAS2OBJS   += $(CDBLAS2OBJS)
 DBLAS3OBJS   += $(CDBLAS3OBJS)
@@ -380,6 +388,7 @@ ZBLAS3OBJS   += $(CZBLAS3OBJS)
 endif
 
 SBLASOBJS    = $(SBLAS1OBJS) $(SBLAS2OBJS) $(SBLAS3OBJS)
+SHBLASOBJS   = $(SHBLAS3OBJS)
 DBLASOBJS    = $(DBLAS1OBJS) $(DBLAS2OBJS) $(DBLAS3OBJS)
 QBLASOBJS    = $(QBLAS1OBJS) $(QBLAS2OBJS) $(QBLAS3OBJS)
 CBLASOBJS    = $(CBLAS1OBJS) $(CBLAS2OBJS) $(CBLAS3OBJS)
@@ -454,7 +463,7 @@ ZBLASOBJS += $(ZLAPACKOBJS)
 
 endif
 
-FUNCOBJS    = $(SBLASOBJS) $(DBLASOBJS) $(CBLASOBJS) $(ZBLASOBJS)
+FUNCOBJS    = $(SHBLASOBJS) $(SBLASOBJS) $(DBLASOBJS) $(CBLASOBJS) $(ZBLASOBJS)
 
 ifdef EXPRECISION
 FUNCOBJS   += $(QBLASOBJS) $(XBLASOBJS)
@@ -488,10 +497,10 @@ level1 : $(SBLAS1OBJS) $(DBLAS1OBJS) $(QBLAS1OBJS) $(CBLAS1OBJS) $(ZBLAS1OBJS) $
 level2 : $(SBLAS2OBJS) $(DBLAS2OBJS) $(QBLAS2OBJS) $(CBLAS2OBJS) $(ZBLAS2OBJS) $(XBLAS2OBJS)
 	$(AR) $(ARFLAGS) -ru $(TOPDIR)/$(LIBNAME) $^
 
-level3 : $(SBLAS3OBJS) $(DBLAS3OBJS) $(QBLAS3OBJS) $(CBLAS3OBJS) $(ZBLAS3OBJS) $(XBLAS3OBJS)
+level3 : $(SHBLAS3OBJS) $(SBLAS3OBJS) $(DBLAS3OBJS) $(QBLAS3OBJS) $(CBLAS3OBJS) $(ZBLAS3OBJS) $(XBLAS3OBJS)
 	$(AR) $(ARFLAGS) -ru $(TOPDIR)/$(LIBNAME) $^
 
-$(CSBLASOBJS) $(CSBLASOBJS_P) $(CDBLASOBJS) $(CDBLASOBJS_P) $(CQBLASOBJS) $(CQBLASOBJS_P) \
+$(CSHBLASOBJS) $(CSHBLASOBJS_P) $(CSBLASOBJS) $(CSBLASOBJS_P) $(CDBLASOBJS) $(CDBLASOBJS_P) $(CQBLASOBJS) $(CQBLASOBJS_P) \
 $(CCBLASOBJS) $(CCBLASOBJS_P) $(CZBLASOBJS) $(CZBLASOBJS_P) $(CXBLASOBJS) $(CXBLASOBJS_P) : override CFLAGS += -DCBLAS
 
 srot.$(SUFFIX) srot.$(PSUFFIX) : rot.c
@@ -1209,6 +1218,11 @@ zhpr2.$(SUFFIX) zhpr2.$(PSUFFIX) : zhpr2.c
 xhpr2.$(SUFFIX) xhpr2.$(PSUFFIX) : zhpr2.c
 	$(CC) -c $(CFLAGS) $< -o $(@F)
 
+ifeq ($(BUILD_HALF),1)
+shgemm.$(SUFFIX) shgemm.$(PSUFFIX) : gemm.c ../param.h
+	$(CC) -c $(CFLAGS) $< -o $(@F)
+endif
+
 sgemm.$(SUFFIX) sgemm.$(PSUFFIX) : gemm.c ../param.h
 	$(CC) -c $(CFLAGS) $< -o $(@F)
 
@@ -1770,6 +1784,11 @@ cblas_zhemv.$(SUFFIX) cblas_zhemv.$(PSUFFIX) : zhemv.c
 cblas_sgemm.$(SUFFIX) cblas_sgemm.$(PSUFFIX) : gemm.c ../param.h
 	$(CC) -DCBLAS -c $(CFLAGS) $< -o $(@F)
 
+ifeq ($(BUILD_HALF),1)
+cblas_shgemm.$(SUFFIX) cblas_shgemm.$(PSUFFIX) : gemm.c ../param.h
+	$(CC) -DCBLAS -c $(CFLAGS) $< -o $(@F)
+endif
+
 cblas_dgemm.$(SUFFIX) cblas_dgemm.$(PSUFFIX) : gemm.c ../param.h
 	$(CC) -DCBLAS -c $(CFLAGS) $< -o $(@F)
 
diff --git a/interface/gemm.c b/interface/gemm.c
index 97e71bc85..99388e7d9 100644
--- a/interface/gemm.c
+++ b/interface/gemm.c
@@ -77,7 +77,7 @@
 #define GEMM_MULTITHREAD_THRESHOLD 4
 #endif
 
-static int (*gemm[])(blas_arg_t *, BLASLONG *, BLASLONG *, FLOAT *, FLOAT *, BLASLONG) = {
+static int (*gemm[])(blas_arg_t *, BLASLONG *, BLASLONG *, IFLOAT *, IFLOAT *, BLASLONG) = {
 #ifndef GEMM3M
   GEMM_NN, GEMM_TN, GEMM_RN, GEMM_CN,
   GEMM_NT, GEMM_TT, GEMM_RT, GEMM_CT,
@@ -108,8 +108,8 @@ static int (*gemm[])(blas_arg_t *, BLASLONG *, BLASLONG *, FLOAT *, FLOAT *, BLA
 void NAME(char *TRANSA, char *TRANSB,
 	  blasint *M, blasint *N, blasint *K,
 	  FLOAT *alpha,
-	  FLOAT *a, blasint *ldA,
-	  FLOAT *b, blasint *ldB,
+	  IFLOAT *a, blasint *ldA,
+	  IFLOAT *b, blasint *ldB,
 	  FLOAT *beta,
 	  FLOAT *c, blasint *ldC){
 
@@ -119,8 +119,8 @@ void NAME(char *TRANSA, char *TRANSB,
   blasint info;
 
   char transA, transB;
-  FLOAT *buffer;
-  FLOAT *sa, *sb;
+  IFLOAT *buffer;
+  IFLOAT *sa, *sb;
 
 #ifdef SMP
   double MNK;
@@ -272,6 +272,9 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_TRANSPOSE TransA, enum CBLAS_TRANS
   PRINT_DEBUG_CNAME;
 
 #if !defined(COMPLEX) && !defined(DOUBLE) && defined(USE_SGEMM_KERNEL_DIRECT)
+#ifdef DYNAMIC_ARCH
+ if (support_avx512() )
+#endif  
   if (beta == 0 && alpha == 1.0 && order == CblasRowMajor && TransA == CblasNoTrans && TransB == CblasNoTrans && sgemm_kernel_direct_performant(m,n,k)) {
 	sgemm_kernel_direct(m, n, k, a, lda, b, ldb, c, ldc);
 	return;
diff --git a/kernel/CMakeLists.txt b/kernel/CMakeLists.txt
index 35e0fff25..b114c6a33 100644
--- a/kernel/CMakeLists.txt
+++ b/kernel/CMakeLists.txt
@@ -41,6 +41,9 @@ function (build_core TARGET_CORE KDIR TSUFFIX KERNEL_DEFINITIONS)
     foreach (float_type ${FLOAT_TYPES})
       # a bit of metaprogramming here to pull out the appropriate KERNEL var
       string(SUBSTRING ${float_type} 0 1 float_char)
+      if (${float_type} STREQUAL "HALF")
+	set (float_char "SH")
+      endif ()
       GenerateNamedObjects("${KERNELDIR}/${${float_char}AMAXKERNEL}" "USE_ABS" "amax_k" false "" "" false ${float_type})
       GenerateNamedObjects("${KERNELDIR}/${${float_char}AMINKERNEL}" "USE_ABS;USE_MIN" "amin_k" false "" "" false ${float_type})
       if (DEFINED ${float_char}MAXKERNEL)
@@ -93,6 +96,9 @@ function (build_core TARGET_CORE KDIR TSUFFIX KERNEL_DEFINITIONS)
     GenerateNamedObjects("generic/ger.c" "" "ger_k" false "" "" "" 3)
     foreach (float_type ${FLOAT_TYPES})
       string(SUBSTRING ${float_type} 0 1 float_char)
+      if (${float_type} STREQUAL "HALF")
+	set (float_char "SH")
+      endif ()
       if (${float_type} STREQUAL "COMPLEX" OR ${float_type} STREQUAL "ZCOMPLEX")
         GenerateNamedObjects("${KERNELDIR}/${${float_char}GERUKERNEL}" "" "geru_k" false "" "" false ${float_type})
         GenerateNamedObjects("${KERNELDIR}/${${float_char}GERCKERNEL}" "CONJ" "gerc_k" false "" "" false ${float_type})
@@ -128,13 +134,23 @@ function (build_core TARGET_CORE KDIR TSUFFIX KERNEL_DEFINITIONS)
       set(USE_TRMM true)
     endif ()
 
-    foreach (float_type SINGLE DOUBLE)
+    foreach (float_type SINGLE DOUBLE HALF)
       string(SUBSTRING ${float_type} 0 1 float_char)
+      if (${float_type} STREQUAL "HALF")
+        if (NOT ${BUILD_HALF})
+	  continue ()
+        else ()
+	  set (float_char "SH")
+      endif ()
+      endif ()
       GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMMKERNEL}" "" "gemm_kernel" false "" "" false ${float_type})
     endforeach()
 
     foreach (float_type ${FLOAT_TYPES})
       string(SUBSTRING ${float_type} 0 1 float_char)
+      if (${float_type} STREQUAL "HALF")
+	set (float_char "SH")
+      endif ()
       if (${float_char}GEMMINCOPY)
         GenerateNamedObjects("${KERNELDIR}/${${float_char}GEMMINCOPY}" "${float_type}" "${${float_char}GEMMINCOPYOBJ}" false "" "" true ${float_type})
       endif ()
@@ -470,9 +486,13 @@ function (build_core TARGET_CORE KDIR TSUFFIX KERNEL_DEFINITIONS)
       GenerateNamedObjects("${KERNELDIR}/${${float_char}GEADD_KERNEL}" "" "geadd_k" false "" "" false ${float_type})
     endforeach ()
 
+
     # Makefile.LA
     if(NOT NO_LAPACK)
       foreach (float_type ${FLOAT_TYPES})
+      if (${float_type} STREQUAL "HALF")
+	set (float_char "SH")
+      endif ()
         if (NOT DEFINED ${float_char}NEG_TCOPY)
           if (${float_char} STREQUAL "Z" OR ${float_char} STREQUAL "C" OR ${float_char} STREQUAL "X")
             set(${float_char}NEG_TCOPY ../generic/zneg_tcopy.c)
@@ -516,6 +536,9 @@ function (build_core TARGET_CORE KDIR TSUFFIX KERNEL_DEFINITIONS)
     foreach (float_type ${FLOAT_TYPES})
       # a bit of metaprogramming here to pull out the appropriate KERNEL var
       string(SUBSTRING ${float_type} 0 1 float_char)
+      if (${float_type} STREQUAL "HALF")
+	set (float_char "SH")
+      endif ()
       GenerateNamedObjects("generic/neg_tcopy_${${float_char}GEMM_UNROLL_M}.c" "" "neg_tcopy" false "" ${TSUFFIX}  false ${float_type})
       GenerateNamedObjects("generic/laswp_ncopy_${${float_char}GEMM_UNROLL_N}.c" "" "laswp_ncopy" false "" ${TSUFFIX}  false ${float_type})
     endforeach ()
diff --git a/kernel/Makefile.L3 b/kernel/Makefile.L3
index 6d96abb2e..da6c5fd57 100644
--- a/kernel/Makefile.L3
+++ b/kernel/Makefile.L3
@@ -59,6 +59,25 @@ ifeq ($(CORE), Z14)
 USE_TRMM = 1
 endif
 
+ifeq ($(BUILD_HALF), 1)
+ifndef SHGEMMKERNEL
+SHGEMM_BETA = ../generic/gemm_beta.c
+SHGEMMKERNEL    = ../generic/gemmkernel_2x2.c
+SHGEMMINCOPY    = ../generic/gemm_ncopy_2.c
+SHGEMMITCOPY    = ../generic/gemm_tcopy_2.c
+SHGEMMONCOPY    = ../generic/gemm_ncopy_2.c
+SHGEMMOTCOPY    = ../generic/gemm_tcopy_2.c
+SHGEMMINCOPYOBJ =  shgemm_incopy$(TSUFFIX).$(SUFFIX)
+SHGEMMITCOPYOBJ =  shgemm_itcopy$(TSUFFIX).$(SUFFIX)
+SHGEMMONCOPYOBJ =  shgemm_oncopy$(TSUFFIX).$(SUFFIX)
+SHGEMMOTCOPYOBJ =  shgemm_otcopy$(TSUFFIX).$(SUFFIX)
+endif
+
+SHKERNELOBJS	+= \
+	shgemm_kernel$(TSUFFIX).$(SUFFIX) \
+	$(SHGEMMINCOPYOBJ) $(SHGEMMITCOPYOBJ) \
+	$(SHGEMMONCOPYOBJ) $(SHGEMMOTCOPYOBJ)
+endif
 
 SKERNELOBJS	+= \
 	sgemm_kernel$(TSUFFIX).$(SUFFIX) \
@@ -93,6 +112,9 @@ XKERNELOBJS	+= \
 	$(XGEMMINCOPYOBJ) $(XGEMMITCOPYOBJ) \
 	$(XGEMMONCOPYOBJ) $(XGEMMOTCOPYOBJ)
 
+ifeq ($(BUILD_HALF),1)
+SHBLASOBJS      += $(SHKERNELOBJS)
+endif
 SBLASOBJS	+= $(SKERNELOBJS)
 DBLASOBJS	+= $(DKERNELOBJS)
 QBLASOBJS	+= $(QKERNELOBJS)
@@ -100,6 +122,10 @@ CBLASOBJS	+= $(CKERNELOBJS)
 ZBLASOBJS	+= $(ZKERNELOBJS)
 XBLASOBJS	+= $(XKERNELOBJS)
 
+ifeq ($(BUILD_HALF),1)
+SHBLASOBJS += shgemm_beta$(TSUFFIX).$(SUFFIX)
+endif
+
 SBLASOBJS	+= \
 	sgemm_beta$(TSUFFIX).$(SUFFIX) \
 	strmm_kernel_LN$(TSUFFIX).$(SUFFIX) strmm_kernel_LT$(TSUFFIX).$(SUFFIX) \
@@ -389,6 +415,12 @@ ZBLASOBJS += \
 	zimatcopy_k_ctc$(TSUFFIX).$(SUFFIX) zimatcopy_k_rtc$(TSUFFIX).$(SUFFIX) \
 	zgeadd_k$(TSUFFIX).$(SUFFIX) 
 
+ifeq ($(BUILD_HALF), 1)
+SHGEMMINCOPYOBJ_P = $(SHGEMMINCOPYOBJ:.$(SUFFIX)=.$(PSUFFIX))
+SHGEMMITCOPYOBJ_P = $(SHGEMMITCOPYOBJ:.$(SUFFIX)=.$(PSUFFIX))
+SHGEMMONCOPYOBJ_P = $(SHGEMMONCOPYOBJ:.$(SUFFIX)=.$(PSUFFIX))
+SHGEMMOTCOPYOBJ_P = $(SHGEMMOTCOPYOBJ:.$(SUFFIX)=.$(PSUFFIX))
+endif
 
 SGEMMINCOPYOBJ_P = $(SGEMMINCOPYOBJ:.$(SUFFIX)=.$(PSUFFIX))
 SGEMMITCOPYOBJ_P = $(SGEMMITCOPYOBJ:.$(SUFFIX)=.$(PSUFFIX))
@@ -415,6 +447,11 @@ XGEMMITCOPYOBJ_P = $(XGEMMITCOPYOBJ:.$(SUFFIX)=.$(PSUFFIX))
 XGEMMONCOPYOBJ_P = $(XGEMMONCOPYOBJ:.$(SUFFIX)=.$(PSUFFIX))
 XGEMMOTCOPYOBJ_P = $(XGEMMOTCOPYOBJ:.$(SUFFIX)=.$(PSUFFIX))
 
+ifeq ($(BUILD_HALF),1)
+$(KDIR)shgemm_beta$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SHGEMM_BETA)
+	$(CC) $(CFLAGS) -c -DHALF -UDOUBLE -UCOMPLEX $< -o $@
+endif
+
 $(KDIR)sgemm_beta$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SGEMM_BETA)
 	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
 
@@ -433,6 +470,41 @@ $(KDIR)zgemm_beta$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(ZGEMM_BETA)
 $(KDIR)xgemm_beta$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(XGEMM_BETA)
 	$(CC) $(CFLAGS) -c -DXDOUBLE -DCOMPLEX $< -o $@
 
+
+ifeq ($(BUILD_HALF), 1)
+
+$(KDIR)$(SHGEMMONCOPYOBJ) : $(KERNELDIR)/$(SHGEMMONCOPY)
+	$(CC) $(CFLAGS) -c -DHALF -UDOUBLE -UCOMPLEX $< -o $@
+
+$(KDIR)$(SHGEMMOTCOPYOBJ) : $(KERNELDIR)/$(SHGEMMOTCOPY)
+
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -E -DHALF -UDOUBLE -UCOMPLEX $< -o shgemmotcopy.s
+	m4 shgemmotcopy.s > shgemmotcopy_nomacros.s
+	$(CC) $(CFLAGS) -c -DHALF -UDOUBLE -UCOMPLEX shgemmotcopy_nomacros.s -o $@
+	rm shgemmotcopy.s shgemmotcopy_nomacros.s
+else
+	$(CC) $(CFLAGS) -c -DHALF -UDOUBLE -UCOMPLEX $< -o $@
+endif
+
+ifneq ($(SHGEMM_UNROLL_M), $(SHGEMM_UNROLL_N))
+
+$(KDIR)$(SHGEMMINCOPYOBJ) : $(KERNELDIR)/$(SHGEMMINCOPY)
+	$(CC) $(CFLAGS) -c -DHALF -UDOUBLE -UCOMPLEX $< -o $@
+
+$(KDIR)$(SHGEMMITCOPYOBJ) : $(KERNELDIR)/$(SHGEMMITCOPY)
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -E -DHALF -UDOUBLE -UCOMPLEX $< -o shgemmitcopy.s
+	m4 shgemmitcopy.s > shgemmitcopy_nomacros.s
+	$(CC) $(CFLAGS) -c -DHALF -UDOUBLE -UCOMPLEX shgemmitcopy_nomacros.s -o $@
+	rm shgemmitcopy.s shgemmitcopy_nomacros.s
+else
+	$(CC) $(CFLAGS) -c -DHALF -UDOUBLE -UCOMPLEX $< -o $@
+endif
+
+endif
+endif
+
 $(KDIR)$(SGEMMONCOPYOBJ) : $(KERNELDIR)/$(SGEMMONCOPY)
 	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
 
@@ -590,6 +662,19 @@ else
 	$(CC) $(CFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
 endif
 
+ifeq ($(BUILD_HALF), 1)
+
+$(KDIR)shgemm_kernel$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SHGEMMKERNEL) $(SHGEMMDEPEND)
+ifeq ($(OS), AIX)
+	$(CC) $(CFLAGS) -E -DHALF -UDOUBLE -UCOMPLEX  $< -o shgemm_kernel$(TSUFFIX).s
+	m4 shgemm_kernel$(TSUFFIX).s > shgemm_kernel$(TSUFFIX)_nomacros.s
+	$(CC) $(CFLAGS) -c -DHALF -UDOUBLE -UCOMPLEX shgemm_kernel$(TSUFFIX)_nomacros.s -o $@
+	rm shgemm_kernel$(TSUFFIX).s shgemm_kernel$(TSUFFIX)_nomacros.s
+else
+	$(CC) $(CFLAGS) -c -DHALF -UDOUBLE -UCOMPLEX $< -o $@
+endif
+endif
+
 $(KDIR)dgemm_kernel$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEMMKERNEL) $(DGEMMDEPEND)
 ifeq ($(OS), AIX)
 	$(CC) $(CFLAGS) -E -DDOUBLE -UCOMPLEX $< -o dgemm_kernel$(TSUFFIX).s
@@ -2206,6 +2291,11 @@ $(KDIR)xtrsm_oltncopy$(TSUFFIX).$(SUFFIX) : generic/ztrsm_ltcopy_$(XGEMM_UNROLL_
 $(KDIR)sgemm_beta$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(SGEMM_BETA)
 	$(CC) $(PFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
 
+ifeq ($(BUILD_HALF),1)
+$(KDIR)shgemm_beta$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(SHGEMM_BETA)
+	$(CC) $(PFLAGS) -c -DHALF -UDOUBLE -UCOMPLEX $< -o $@
+endif
+
 $(KDIR)dgemm_beta$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(DGEMM_BETA)
 	$(CC) $(PFLAGS) -c -DDOUBLE -UCOMPLEX $< -o $@
 
@@ -2221,6 +2311,24 @@ $(KDIR)zgemm_beta$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(ZGEMM_BETA)
 $(KDIR)xgemm_beta$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(XGEMM_BETA)
 	$(CC) $(PFLAGS) -c -DXDOUBLE -DCOMPLEX $< -o $@
 
+
+ifeq ($(BUILD_HALF), 1)
+$(SHGEMMONCOPYOBJ_P) : $(KERNELDIR)/$(SHGEMMONCOPY)
+	$(CC) $(PFLAGS) -c -DHALF -UDOUBLE -UCOMPLEX $< -o $@
+
+$(SHGEMMOTCOPYOBJ_P) : $(KERNELDIR)/$(SHGEMMOTCOPY)
+	$(CC) $(PFLAGS) -c -DHALF -UDOUBLE -UCOMPLEX $< -o $@
+
+ifneq ($(SHGEMM_UNROLL_M), $(SHGEMM_UNROLL_N))
+$(SHGEMMINCOPYOBJ_P) : $(KERNELDIR)/$(SHGEMMINCOPY)
+	$(CC) $(PFLAGS) -c -DHALF -UDOUBLE -UCOMPLEX $< -o $@
+
+$(SHGEMMITCOPYOBJ_P) : $(KERNELDIR)/$(SHGEMMITCOPY)
+	$(CC) $(PFLAGS) -c -DHALF -UDOUBLE -UCOMPLEX $< -o $@
+
+endif
+endif
+
 $(SGEMMONCOPYOBJ_P) : $(KERNELDIR)/$(SGEMMONCOPY)
 	$(CC) $(PFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
 
@@ -2325,6 +2433,12 @@ endif
 
 endif
 
+
+ifeq ($(BUILD_HALF), 1)
+$(KDIR)shgemm_kernel$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(SHGEMMKERNEL) $(SHGEMMDEPEND)
+	$(CC) $(PFLAGS) -c -DHALF -UDOUBLE -UCOMPLEX $< -o $@
+endif
+
 $(KDIR)sgemm_kernel$(TSUFFIX).$(PSUFFIX) : $(KERNELDIR)/$(SGEMMKERNEL) $(SGEMMDEPEND)
 	$(CC) $(PFLAGS) -c -UDOUBLE -UCOMPLEX $< -o $@
 
diff --git a/kernel/arm/nrm2_vfpv3.S b/kernel/arm/nrm2_vfpv3.S
index 7be1e977e..82ae5e8d4 100644
--- a/kernel/arm/nrm2_vfpv3.S
+++ b/kernel/arm/nrm2_vfpv3.S
@@ -61,20 +61,20 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	vldmia.f64	X!, 	{ d4 }
 	vcmpe.f64	d4, d6				// compare with 0.0
 	vmrs		APSR_nzcv, fpscr
-	beq		KERNEL_F1_NEXT_\@
+	beq		1f     /* KERNEL_F1_NEXT_\@ */
 	vabs.f64   	d4,  d4
 	vcmpe.f64  	d0,  d4				// compare with scale
 	vmrs		APSR_nzcv, fpscr
 	vdivge.f64	d2 , d4, d0			// scale >= x ?	x / scale
 	vmlage.f64	d1 , d2 , d2			// ssq += ( x/scale ) * ( x/scale )
-	bge		KERNEL_F1_NEXT_\@
+	bge		1f     /* KERNEL_F1_NEXT_\@ */
 	vdiv.f64	d2 , d0, d4			// scale / x
 	vmul.f64	d2 , d2, d2			// ( scale / x ) * ( scale / x )
 	vmul.f64	d3 , d1, d2			// ssq * ( scale / x ) * ( scale / x )
 	vadd.f64	d1 , d3, d7			// ssq = 1 + ssq * ( scale / x ) * ( scale / x )
 	vmov.f64	d0 , d4				// scale = x
 
-KERNEL_F1_NEXT_\@:
+1:                             /* KERNEL_F1_NEXT_\@: */
 
 .endm
 
@@ -124,20 +124,20 @@ KERNEL_S1_NEXT:
 	vldmia.f32	X!, 	{ s4 }
 	vcmpe.f32	s4, s6				// compare with 0.0
 	vmrs		APSR_nzcv, fpscr
-	beq		KERNEL_F1_NEXT_\@
+	beq		1f        /* KERNEL_F1_NEXT_\@ */
 	vabs.f32   	s4,  s4
 	vcmpe.f32  	s0,  s4				// compare with scale
 	vmrs		APSR_nzcv, fpscr
 	vdivge.f32	s2 , s4, s0			// scale >= x ?	x / scale
 	vmlage.f32	s1 , s2 , s2			// ssq += ( x/scale ) * ( x/scale )
-	bge		KERNEL_F1_NEXT_\@
+	bge		1f        /* KERNEL_F1_NEXT_\@ */
 	vdiv.f32	s2 , s0, s4			// scale / x
 	vmul.f32	s2 , s2, s2			// ( scale / x ) * ( scale / x )
 	vmul.f32	s3 , s1, s2			// ssq * ( scale / x ) * ( scale / x )
 	vadd.f32	s1 , s3, s7			// ssq = 1 + ssq * ( scale / x ) * ( scale / x )
 	vmov.f32	s0 , s4				// scale = x
 
-KERNEL_F1_NEXT_\@:
+1:                                /* KERNEL_F1_NEXT_\@: */
 
 .endm
 
@@ -195,37 +195,37 @@ KERNEL_S1_NEXT:
 
 	vcmpe.f64	d4, d6				// compare with 0.0
 	vmrs		APSR_nzcv, fpscr
-	beq		KERNEL_F1_NEXT_\@
+	beq		1f           /* KERNEL_F1_NEXT_\@ */
 	vabs.f64   	d4,  d4
 	vcmpe.f64  	d0,  d4				// compare with scale
 	vmrs		APSR_nzcv, fpscr
 	vdivge.f64	d2 , d4, d0			// scale >= x ?	x / scale
 	vmlage.f64	d1 , d2 , d2			// ssq += ( x/scale ) * ( x/scale )
-	bge		KERNEL_F1_NEXT_\@
+	bge		1f           /* KERNEL_F1_NEXT_\@ */
 	vdiv.f64	d2 , d0, d4			// scale / x
 	vmul.f64	d2 , d2, d2			// ( scale / x ) * ( scale / x )
 	vmul.f64	d3 , d1, d2			// ssq * ( scale / x ) * ( scale / x )
 	vadd.f64	d1 , d3, d7			// ssq = 1 + ssq * ( scale / x ) * ( scale / x )
 	vmov.f64	d0 , d4				// scale = x
 
-KERNEL_F1_NEXT_\@:
+1:                                   /* KERNEL_F1_NEXT_\@: */
 
 	vcmpe.f64	d5, d6				// compare with 0.0
 	vmrs		APSR_nzcv, fpscr
-	beq		KERNEL_F1_END_\@
+	beq		2f           /* KERNEL_F1_END_\@ */
 	vabs.f64   	d5,  d5
 	vcmpe.f64  	d0,  d5				// compare with scale
 	vmrs		APSR_nzcv, fpscr
 	vdivge.f64	d2 , d5, d0			// scale >= x ?	x / scale
 	vmlage.f64	d1 , d2 , d2			// ssq += ( x/scale ) * ( x/scale )
-	bge		KERNEL_F1_END_\@
+	bge		2f           /* KERNEL_F1_END_\@ */
 	vdiv.f64	d2 , d0, d5			// scale / x
 	vmul.f64	d2 , d2, d2			// ( scale / x ) * ( scale / x )
 	vmul.f64	d3 , d1, d2			// ssq * ( scale / x ) * ( scale / x )
 	vadd.f64	d1 , d3, d7			// ssq = 1 + ssq * ( scale / x ) * ( scale / x )
 	vmov.f64	d0 , d5				// scale = x
 
-KERNEL_F1_END_\@:
+2:                                   /* KERNEL_F1_END_\@: */
 
 
 .endm
@@ -253,37 +253,37 @@ KERNEL_F1_END_\@:
 
 	vcmpe.f64	d4, d6				// compare with 0.0
 	vmrs		APSR_nzcv, fpscr
-	beq		KERNEL_S1_NEXT_\@
+	beq		1f          /* KERNEL_S1_NEXT_\@ */
 	vabs.f64   	d4,  d4
 	vcmpe.f64  	d0,  d4				// compare with scale
 	vmrs		APSR_nzcv, fpscr
 	vdivge.f64	d2 , d4, d0			// scale >= x ?	x / scale
 	vmlage.f64	d1 , d2 , d2			// ssq += ( x/scale ) * ( x/scale )
-	bge		KERNEL_S1_NEXT_\@
+	bge		1f          /* KERNEL_S1_NEXT_\@ */
 	vdiv.f64	d2 , d0, d4			// scale / x
 	vmul.f64	d2 , d2, d2			// ( scale / x ) * ( scale / x )
 	vmul.f64	d3 , d1, d2			// ssq * ( scale / x ) * ( scale / x )
 	vadd.f64	d1 , d3, d7			// ssq = 1 + ssq * ( scale / x ) * ( scale / x )
 	vmov.f64	d0 , d4				// scale = x
 
-KERNEL_S1_NEXT_\@:
+1:                                  /* KERNEL_S1_NEXT_\@: */
 
 	vcmpe.f64	d5, d6				// compare with 0.0
 	vmrs		APSR_nzcv, fpscr
-	beq		KERNEL_S1_END_\@
+	beq		2f          /* KERNEL_S1_END_\@ */
 	vabs.f64   	d5,  d5
 	vcmpe.f64  	d0,  d5				// compare with scale
 	vmrs		APSR_nzcv, fpscr
 	vdivge.f64	d2 , d5, d0			// scale >= x ?	x / scale
 	vmlage.f64	d1 , d2 , d2			// ssq += ( x/scale ) * ( x/scale )
-	bge		KERNEL_S1_END_\@
+	bge		2f          /* KERNEL_S1_END_\@ */
 	vdiv.f64	d2 , d0, d5			// scale / x
 	vmul.f64	d2 , d2, d2			// ( scale / x ) * ( scale / x )
 	vmul.f64	d3 , d1, d2			// ssq * ( scale / x ) * ( scale / x )
 	vadd.f64	d1 , d3, d7			// ssq = 1 + ssq * ( scale / x ) * ( scale / x )
 	vmov.f64	d0 , d5				// scale = x
 
-KERNEL_S1_END_\@:
+2:                                  /* KERNEL_S1_END_\@: */
 
 	add	X, X, INC_X
 
@@ -298,37 +298,37 @@ KERNEL_S1_END_\@:
 
 	vcmpe.f32	s4, s6				// compare with 0.0
 	vmrs		APSR_nzcv, fpscr
-	beq		KERNEL_F1_NEXT_\@
+	beq		1f         /* KERNEL_F1_NEXT_\@ */
 	vabs.f32   	s4,  s4
 	vcmpe.f32  	s0,  s4				// compare with scale
 	vmrs		APSR_nzcv, fpscr
 	vdivge.f32	s2 , s4, s0			// scale >= x ?	x / scale
 	vmlage.f32	s1 , s2 , s2			// ssq += ( x/scale ) * ( x/scale )
-	bge		KERNEL_F1_NEXT_\@
+	bge		1f         /* KERNEL_F1_NEXT_\@ */
 	vdiv.f32	s2 , s0, s4			// scale / x
 	vmul.f32	s2 , s2, s2			// ( scale / x ) * ( scale / x )
 	vmul.f32	s3 , s1, s2			// ssq * ( scale / x ) * ( scale / x )
 	vadd.f32	s1 , s3, s7			// ssq = 1 + ssq * ( scale / x ) * ( scale / x )
 	vmov.f32	s0 , s4				// scale = x
 
-KERNEL_F1_NEXT_\@:
+1:                                 /* KERNEL_F1_NEXT_\@: */
 
 	vcmpe.f32	s5, s6				// compare with 0.0
 	vmrs		APSR_nzcv, fpscr
-	beq		KERNEL_F1_END_\@
+	beq		2f         /* KERNEL_F1_END_\@ */
 	vabs.f32   	s5,  s5
 	vcmpe.f32  	s0,  s5				// compare with scale
 	vmrs		APSR_nzcv, fpscr
 	vdivge.f32	s2 , s5, s0			// scale >= x ?	x / scale
 	vmlage.f32	s1 , s2 , s2			// ssq += ( x/scale ) * ( x/scale )
-	bge		KERNEL_F1_END_\@
+	bge		2f         /* KERNEL_F1_END_\@ */
 	vdiv.f32	s2 , s0, s5			// scale / x
 	vmul.f32	s2 , s2, s2			// ( scale / x ) * ( scale / x )
 	vmul.f32	s3 , s1, s2			// ssq * ( scale / x ) * ( scale / x )
 	vadd.f32	s1 , s3, s7			// ssq = 1 + ssq * ( scale / x ) * ( scale / x )
 	vmov.f32	s0 , s5				// scale = x
 
-KERNEL_F1_END_\@:
+2:                                 /* KERNEL_F1_END_\@: */
 
 
 .endm
@@ -354,37 +354,37 @@ KERNEL_F1_END_\@:
 
 	vcmpe.f32	s4, s6				// compare with 0.0
 	vmrs		APSR_nzcv, fpscr
-	beq		KERNEL_S1_NEXT_\@
+	beq		1f          /* KERNEL_S1_NEXT_\@ */
 	vabs.f32   	s4,  s4
 	vcmpe.f32  	s0,  s4				// compare with scale
 	vmrs		APSR_nzcv, fpscr
 	vdivge.f32	s2 , s4, s0			// scale >= x ?	x / scale
 	vmlage.f32	s1 , s2 , s2			// ssq += ( x/scale ) * ( x/scale )
-	bge		KERNEL_S1_NEXT_\@
+	bge		1f          /* KERNEL_S1_NEXT_\@ */
 	vdiv.f32	s2 , s0, s4			// scale / x
 	vmul.f32	s2 , s2, s2			// ( scale / x ) * ( scale / x )
 	vmul.f32	s3 , s1, s2			// ssq * ( scale / x ) * ( scale / x )
 	vadd.f32	s1 , s3, s7			// ssq = 1 + ssq * ( scale / x ) * ( scale / x )
 	vmov.f32	s0 , s4				// scale = x
 
-KERNEL_S1_NEXT_\@:
+1:                                  /* KERNEL_S1_NEXT_\@: */
 
 	vcmpe.f32	s5, s6				// compare with 0.0
 	vmrs		APSR_nzcv, fpscr
-	beq		KERNEL_S1_END_\@
+	beq		2f          /* KERNEL_S1_END_\@ */
 	vabs.f32   	s5,  s5
 	vcmpe.f32  	s0,  s5				// compare with scale
 	vmrs		APSR_nzcv, fpscr
 	vdivge.f32	s2 , s5, s0			// scale >= x ?	x / scale
 	vmlage.f32	s1 , s2 , s2			// ssq += ( x/scale ) * ( x/scale )
-	bge		KERNEL_S1_END_\@
+	bge		2f          /* KERNEL_S1_END_\@ */
 	vdiv.f32	s2 , s0, s5			// scale / x
 	vmul.f32	s2 , s2, s2			// ( scale / x ) * ( scale / x )
 	vmul.f32	s3 , s1, s2			// ssq * ( scale / x ) * ( scale / x )
 	vadd.f32	s1 , s3, s7			// ssq = 1 + ssq * ( scale / x ) * ( scale / x )
 	vmov.f32	s0 , s5				// scale = x
 
-KERNEL_S1_END_\@:
+2:                                  /* KERNEL_S1_END_\@: */
 
 	add	X, X, INC_X
 
diff --git a/kernel/arm64/KERNEL.CORTEXA53 b/kernel/arm64/KERNEL.CORTEXA53
index c1d33fa3e..eba38a92e 100644
--- a/kernel/arm64/KERNEL.CORTEXA53
+++ b/kernel/arm64/KERNEL.CORTEXA53
@@ -1,3 +1,187 @@
-include $(KERNELDIR)/KERNEL.ARMV8
+SAMINKERNEL  = ../arm/amin.c
+DAMINKERNEL  = ../arm/amin.c
+CAMINKERNEL  = ../arm/zamin.c
+ZAMINKERNEL  = ../arm/zamin.c
+
+SMAXKERNEL   = ../arm/max.c
+DMAXKERNEL   = ../arm/max.c
+
+SMINKERNEL   = ../arm/min.c
+DMINKERNEL   = ../arm/min.c
+
+ISAMINKERNEL = ../arm/iamin.c
+IDAMINKERNEL = ../arm/iamin.c
+ICAMINKERNEL = ../arm/izamin.c
+IZAMINKERNEL = ../arm/izamin.c
+
+ISMAXKERNEL  = ../arm/imax.c
+IDMAXKERNEL  = ../arm/imax.c
+
+ISMINKERNEL  = ../arm/imin.c
+IDMINKERNEL  = ../arm/imin.c
+
+STRSMKERNEL_LN	=  ../generic/trsm_kernel_LN.c
+STRSMKERNEL_LT	=  ../generic/trsm_kernel_LT.c
+STRSMKERNEL_RN	=  ../generic/trsm_kernel_RN.c
+STRSMKERNEL_RT	=  ../generic/trsm_kernel_RT.c
+
+DTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
+DTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
+DTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
+DTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
+
+CTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
+CTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
+CTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
+CTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
+
+ZTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
+ZTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
+ZTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
+ZTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
+
+SAMAXKERNEL  = amax.S
+DAMAXKERNEL  = amax.S
+CAMAXKERNEL  = zamax.S
+ZAMAXKERNEL  = zamax.S
+
+SAXPYKERNEL  = axpy.S
+DAXPYKERNEL  = axpy.S
+CAXPYKERNEL  = zaxpy.S
+ZAXPYKERNEL  = zaxpy.S
+
+SROTKERNEL   = rot.S
+DROTKERNEL   = rot.S
+CROTKERNEL   = zrot.S
+ZROTKERNEL   = zrot.S
+
+SSCALKERNEL  = scal.S
+DSCALKERNEL  = scal.S
+CSCALKERNEL  = zscal.S
+ZSCALKERNEL  = zscal.S
+
+SGEMVNKERNEL = gemv_n.S
+DGEMVNKERNEL = gemv_n.S
+CGEMVNKERNEL = zgemv_n.S
+ZGEMVNKERNEL = zgemv_n.S
+
+SGEMVTKERNEL = gemv_t.S
+DGEMVTKERNEL = gemv_t.S
+CGEMVTKERNEL = zgemv_t.S
+ZGEMVTKERNEL = zgemv_t.S
 
 
+SASUMKERNEL    = asum.S
+DASUMKERNEL    = asum.S
+CASUMKERNEL    = casum.S
+ZASUMKERNEL    = zasum.S
+
+SCOPYKERNEL    = copy.S
+DCOPYKERNEL    = copy.S
+CCOPYKERNEL    = copy.S
+ZCOPYKERNEL    = copy.S
+
+SSWAPKERNEL    = swap.S
+DSWAPKERNEL    = swap.S
+CSWAPKERNEL    = swap.S
+ZSWAPKERNEL    = swap.S
+
+ISAMAXKERNEL   = iamax.S
+IDAMAXKERNEL   = iamax.S
+ICAMAXKERNEL   = izamax.S
+IZAMAXKERNEL   = izamax.S
+
+SNRM2KERNEL    = nrm2.S
+DNRM2KERNEL    = nrm2.S
+CNRM2KERNEL    = znrm2.S
+ZNRM2KERNEL    = znrm2.S
+
+DDOTKERNEL     = dot.S
+SDOTKERNEL     = dot.S
+CDOTKERNEL     = zdot.S
+ZDOTKERNEL     = zdot.S
+DSDOTKERNEL    = dot.S
+
+DGEMM_BETA     = dgemm_beta.S
+SGEMM_BETA     = sgemm_beta.S
+
+ifeq ($(SGEMM_UNROLL_M)x$(SGEMM_UNROLL_N), 8x8)
+SGEMMKERNEL    =  sgemm_kernel_$(SGEMM_UNROLL_M)x$(SGEMM_UNROLL_N)_cortexa53.S
+STRMMKERNEL    =  strmm_kernel_$(SGEMM_UNROLL_M)x$(SGEMM_UNROLL_N)_cortexa53.S
+else
+SGEMMKERNEL    =  sgemm_kernel_$(SGEMM_UNROLL_M)x$(SGEMM_UNROLL_N).S
+STRMMKERNEL    =  strmm_kernel_$(SGEMM_UNROLL_M)x$(SGEMM_UNROLL_N).S
+endif
+ifneq ($(SGEMM_UNROLL_M), $(SGEMM_UNROLL_N))
+ifeq ($(SGEMM_UNROLL_M), 16)
+SGEMMITCOPY    =  sgemm_tcopy_$(SGEMM_UNROLL_M).S
+else
+SGEMMITCOPY    =  ../generic/gemm_tcopy_$(SGEMM_UNROLL_M).c
+endif
+ifeq ($(SGEMM_UNROLL_M), 4)
+SGEMMINCOPY    =  sgemm_ncopy_$(SGEMM_UNROLL_M).S
+else
+SGEMMINCOPY    =  ../generic/gemm_ncopy_$(SGEMM_UNROLL_M).c
+endif
+SGEMMINCOPYOBJ =  sgemm_incopy$(TSUFFIX).$(SUFFIX)
+SGEMMITCOPYOBJ =  sgemm_itcopy$(TSUFFIX).$(SUFFIX)
+endif
+
+SGEMMOTCOPY    =  sgemm_tcopy_$(SGEMM_UNROLL_N).S
+SGEMMONCOPY    =  sgemm_ncopy_$(SGEMM_UNROLL_N).S
+SGEMMONCOPYOBJ =  sgemm_oncopy$(TSUFFIX).$(SUFFIX)
+SGEMMOTCOPYOBJ =  sgemm_otcopy$(TSUFFIX).$(SUFFIX)
+
+DGEMMKERNEL    =  dgemm_kernel_$(DGEMM_UNROLL_M)x$(DGEMM_UNROLL_N).S
+DTRMMKERNEL    =  dtrmm_kernel_$(DGEMM_UNROLL_M)x$(DGEMM_UNROLL_N).S
+
+ifneq ($(DGEMM_UNROLL_M), $(DGEMM_UNROLL_N))
+
+ifeq ($(DGEMM_UNROLL_M), 8)
+DGEMMINCOPY    =  dgemm_ncopy_$(DGEMM_UNROLL_M).S
+DGEMMITCOPY    =  dgemm_tcopy_$(DGEMM_UNROLL_M).S
+else
+DGEMMINCOPY    =  ../generic/gemm_ncopy_$(DGEMM_UNROLL_M).c
+DGEMMITCOPY    =  ../generic/gemm_tcopy_$(DGEMM_UNROLL_M).c
+endif
+
+DGEMMINCOPYOBJ =  dgemm_incopy$(TSUFFIX).$(SUFFIX)
+DGEMMITCOPYOBJ =  dgemm_itcopy$(TSUFFIX).$(SUFFIX)
+endif
+
+ifeq ($(DGEMM_UNROLL_N), 4)
+DGEMMONCOPY    =  dgemm_ncopy_$(DGEMM_UNROLL_N).S
+DGEMMOTCOPY    =  dgemm_tcopy_$(DGEMM_UNROLL_N).S
+else
+DGEMMONCOPY    =  ../generic/gemm_ncopy_$(DGEMM_UNROLL_N).c
+DGEMMOTCOPY    =  ../generic/gemm_tcopy_$(DGEMM_UNROLL_N).c
+endif
+
+DGEMMONCOPYOBJ =  dgemm_oncopy$(TSUFFIX).$(SUFFIX)
+DGEMMOTCOPYOBJ =  dgemm_otcopy$(TSUFFIX).$(SUFFIX)
+
+CGEMMKERNEL    =  cgemm_kernel_$(CGEMM_UNROLL_M)x$(CGEMM_UNROLL_N).S
+CTRMMKERNEL    =  ctrmm_kernel_$(CGEMM_UNROLL_M)x$(CGEMM_UNROLL_N).S
+ifneq ($(CGEMM_UNROLL_M), $(CGEMM_UNROLL_N))
+CGEMMINCOPY    =  ../generic/zgemm_ncopy_$(CGEMM_UNROLL_M).c
+CGEMMITCOPY    =  ../generic/zgemm_tcopy_$(CGEMM_UNROLL_M).c
+CGEMMINCOPYOBJ =  cgemm_incopy$(TSUFFIX).$(SUFFIX)
+CGEMMITCOPYOBJ =  cgemm_itcopy$(TSUFFIX).$(SUFFIX)
+endif
+CGEMMONCOPY    =  ../generic/zgemm_ncopy_$(CGEMM_UNROLL_N).c
+CGEMMOTCOPY    =  ../generic/zgemm_tcopy_$(CGEMM_UNROLL_N).c
+CGEMMONCOPYOBJ =  cgemm_oncopy$(TSUFFIX).$(SUFFIX)
+CGEMMOTCOPYOBJ =  cgemm_otcopy$(TSUFFIX).$(SUFFIX)
+
+ZGEMMKERNEL    =  zgemm_kernel_$(ZGEMM_UNROLL_M)x$(ZGEMM_UNROLL_N).S
+ZTRMMKERNEL    =  ztrmm_kernel_$(ZGEMM_UNROLL_M)x$(ZGEMM_UNROLL_N).S
+ifneq ($(ZGEMM_UNROLL_M), $(ZGEMM_UNROLL_N))
+ZGEMMINCOPY    =  ../generic/zgemm_ncopy_$(ZGEMM_UNROLL_M).c
+ZGEMMITCOPY    =  ../generic/zgemm_tcopy_$(ZGEMM_UNROLL_M).c
+ZGEMMINCOPYOBJ =  zgemm_incopy$(TSUFFIX).$(SUFFIX)
+ZGEMMITCOPYOBJ =  zgemm_itcopy$(TSUFFIX).$(SUFFIX)
+endif
+ZGEMMONCOPY    =  ../generic/zgemm_ncopy_$(ZGEMM_UNROLL_N).c
+ZGEMMOTCOPY    =  ../generic/zgemm_tcopy_$(ZGEMM_UNROLL_N).c
+ZGEMMONCOPYOBJ =  zgemm_oncopy$(TSUFFIX).$(SUFFIX)
+ZGEMMOTCOPYOBJ =  zgemm_otcopy$(TSUFFIX).$(SUFFIX)
diff --git a/kernel/arm64/daxpy_thunderx2t99.S b/kernel/arm64/daxpy_thunderx2t99.S
index b8d0af5c2..baf39150f 100644
--- a/kernel/arm64/daxpy_thunderx2t99.S
+++ b/kernel/arm64/daxpy_thunderx2t99.S
@@ -98,11 +98,58 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	add	X, X, #128
 .endm
 
+/*
+ * No need to do software prefetches if the vector fits
+ * into L1 cache
+ */
+.macro KERNEL_F16_L1CACHE
+	ldp	q4, q5, [X]
+	ldp	q16, q17, [Y]
+
+	ldp	q6, q7, [X, #32]
+	ldp	q18, q19, [Y, #32]
+
+	fmla	v16.2d, v4.2d, v0.d[0]
+	fmla	v17.2d, v5.2d, v0.d[0]
+
+	stp	q16, q17, [Y]
+
+	ldp	q20, q21, [X, #64]
+	ldp	q24, q25, [Y, #64]
+
+	fmla	v18.2d, v6.2d, v0.d[0]
+	fmla	v19.2d, v7.2d, v0.d[0]
+
+	stp	q18, q19, [Y, #32]
+
+	ldp	q22, q23, [X, #96]
+	ldp	q26, q27, [Y, #96]
+
+	fmla	v24.2d, v20.2d, v0.d[0]
+	fmla	v25.2d, v21.2d, v0.d[0]
+
+	stp	q24, q25, [Y, #64]
+
+	fmla	v26.2d, v22.2d, v0.d[0]
+	fmla	v27.2d, v23.2d, v0.d[0]
+
+	stp	q26, q27, [Y, #96]
+
+	add	Y, Y, #128
+	add	X, X, #128
+.endm
+
 .macro KERNEL_F32
 	KERNEL_F16 
 	KERNEL_F16 
 .endm
 
+
+.macro KERNEL_F32_L1CACHE
+	KERNEL_F16_L1CACHE
+	KERNEL_F16_L1CACHE
+.endm
+
 .macro INIT_S
 	lsl	INC_X, INC_X, #3
 	lsl	INC_Y, INC_Y, #3
@@ -138,6 +185,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	cmp	I, xzr
 	beq	.Ldaxpy_kernel_F1
 
+	cmp	N, #2048
+	ble	.Ldaxpy_kernel_F32_L1CACHE
+
 	.align 5
 .Ldaxpy_kernel_F32:
 
@@ -145,6 +195,15 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 	subs	I, I, #1
 	bne	.Ldaxpy_kernel_F32
+	b	.Ldaxpy_kernel_F1
+
+	.align 5
+.Ldaxpy_kernel_F32_L1CACHE:
+
+	KERNEL_F32_L1CACHE
+
+	subs	I, I, #1
+	bne	.Ldaxpy_kernel_F32_L1CACHE
 
 .Ldaxpy_kernel_F1:
 
diff --git a/kernel/arm64/dgemm_beta.S b/kernel/arm64/dgemm_beta.S
index 20011c343..7d21525c2 100644
--- a/kernel/arm64/dgemm_beta.S
+++ b/kernel/arm64/dgemm_beta.S
@@ -81,14 +81,14 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 .endm
 
 .macro INIT_ZERO
-	fmul v0.2d, v0.2d, betaV0
-	fmul v1.2d, v1.2d, betaV0
-	fmul v2.2d, v2.2d, betaV0
-	fmul v3.2d, v3.2d, betaV0
-	fmul v4.2d, v4.2d, betaV0
-	fmul v5.2d, v5.2d, betaV0
-	fmul v6.2d, v6.2d, betaV0
-	fmul v7.2d, v7.2d, betaV0
+	movi v0.2d, #0000000000000000
+	movi v1.2d, #0000000000000000
+	movi v2.2d, #0000000000000000
+	movi v3.2d, #0000000000000000
+	movi v4.2d, #0000000000000000
+	movi v5.2d, #0000000000000000
+	movi v6.2d, #0000000000000000
+	movi v7.2d, #0000000000000000
 .endm
 
 /**************************************************************************************
diff --git a/kernel/arm64/sgemm_beta.S b/kernel/arm64/sgemm_beta.S
index a3b97e231..574485bc4 100755
--- a/kernel/arm64/sgemm_beta.S
+++ b/kernel/arm64/sgemm_beta.S
@@ -81,14 +81,14 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 .endm
 
 .macro INIT_ZERO
-	fmul v0.4s, v0.4s, betaV0
-	fmul v1.4s, v1.4s, betaV0
-	fmul v2.4s, v2.4s, betaV0
-	fmul v3.4s, v3.4s, betaV0
-	fmul v4.4s, v4.4s, betaV0
-	fmul v5.4s, v5.4s, betaV0
-	fmul v6.4s, v6.4s, betaV0
-	fmul v7.4s, v7.4s, betaV0
+	movi v0.4s, #0x0
+	movi v1.4s, #0x0
+	movi v2.4s, #0x0
+	movi v3.4s, #0x0
+	movi v4.4s, #0x0
+	movi v5.4s, #0x0
+	movi v6.4s, #0x0
+	movi v7.4s, #0x0
 .endm
 
 /**************************************************************************************
diff --git a/kernel/arm64/sgemm_kernel_8x8_cortexa53.S b/kernel/arm64/sgemm_kernel_8x8_cortexa53.S
new file mode 100644
index 000000000..628a928ca
--- /dev/null
+++ b/kernel/arm64/sgemm_kernel_8x8_cortexa53.S
@@ -0,0 +1,2299 @@
+/*******************************************************************************
+Copyright (c) 2015, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*******************************************************************************/
+#define ASSEMBLER
+#include "common.h"
+
+/*                   X0          X1          X2          s0        X3        x4       x5           x6  */
+/*int CNAME(BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alpha,FLOAT* ba,FLOAT* bb,FLOAT* C,BLASLONG ldc) */
+
+#define origM		x0
+#define origN		x1
+#define origK		x2
+#define origPA		x3
+#define origPB		x4
+#define pC		x5
+#define LDC		x6
+#define offset		x7
+#define counterL	x8
+#define counterI	x9
+#define counterJ	x10
+#define pB		x11
+#define pCRow0		x12
+#define pCRow1		x13
+#define pCRow2		x14
+#define pA		x15
+#define temp		x16
+
+#define alpha0		s10
+#define alphaV0		v10.s[0]
+#define alpha1		s11
+#define alphaV1		v11.s[0]
+#define alpha2		s14
+#define alphaV2		v14.s[0]
+#define alpha3		s15
+#define alphaV3		v15.s[0]
+
+#define A_PRE_SIZE	640
+#define B_PRE_SIZE	224
+#define C_PRE_SIZE	96
+
+// 00 origM
+// 01 origN
+// 02 origK
+// 03 origPA
+// 04 origPB
+// 05 pC
+// 06 origLDC -> LDC
+// 07 offset
+// 08 counterL
+// 09 counterI
+// 10 counterJ
+// 11 pB
+// 12 pCRow0
+// 13 pCRow1
+// 14 pCRow2
+// 15 pA
+// 16 temp
+// 17
+// 18 must save
+// 19 must save
+// 20 must save pA0_2, pA0_3
+// 21 must save pA0_6, pA0_7
+// 22 must save pA1_2, pA1_3
+// 23 must save pA1_6, pA1_7
+// 24 must save pB0_2, pB0_3
+// 25 must save pB0_6, pB0_7
+// 26 must save pB1_2, pB1_3
+// 27 must save pB1_6, pB1_7
+// 28 must save
+// 29 frame
+// 30 link
+// 31 sp
+
+//v00 ALPHA -> pA0_0, pA0_1, pA0_2, pA0_3
+//v01 pA0_4, pA0_5, pA0_6, pA0_7
+//v02 pA1_0, pA1_1, pA1_2, pA1_3
+//v03 pA1_4, pA1_5, pA1_6, pA1_7
+//v04 pB0_0, pB0_1, pB0_2, pB0_3
+//v05 pB0_4, pB0_5, pB0_6, pB0_7
+//v06 pB1_0, pB1_1, pB1_2, pB1_3
+//v07 pB1_4, pB1_5, pB1_6, pB1_7
+//v08 must save
+//v09 must save
+//v10 must save ALPHA0
+//v11 must save ALPHA1
+//v12 must save
+//v13 must save
+//v14 must save ALPHA2
+//v15 must save ALPHA3
+//v16 must save C00, C01, C02, C03
+//v17 must save C04, C05, C06, C07
+//v18 C08, C09, C10, C11
+//v19 C12, C13, C14, C15
+//v20 C16, C17, C18, C19
+//v21 C20, C21, C22, C23
+//v22 C24, C25, C26, C27
+//v23 C28, C29, C30, C31
+//v24 C32, C33, C34, C35
+//v25 C36, C37, C38, C39
+//v26 C40, C41, C42, C43
+//v27 C44, C45, C46, C47
+//v28 C48, C49, C50, C51
+//v29 C52, C53, C54, C55
+//v30 C56, C57, C58, C59
+//v31 C60, C61, C62, C63
+
+/*******************************************************************************
+* Macro definitions
+*******************************************************************************/
+
+.macro INIT8x8
+	fmov		s16, wzr
+	fmov		s17, wzr
+	fmov		s18, s16
+	fmov		s19, s17
+	fmov		s20, wzr
+	fmov		s21, s16
+	fmov		s22, s17
+	fmov		s23, s18
+	fmov		s24, wzr
+	fmov		s25, s16
+	fmov		s26, s17
+	fmov		s27, s18
+	fmov		s28, wzr
+	fmov		s29, s16
+	fmov		s30, s17
+	fmov		s31, s18
+.endm
+
+.macro KERNEL8x8_I
+	ldp	q0, q1, [pA], #32
+	ldp	q4, q5, [pB], #32
+
+	ldr	d2, [pA], #8
+	ldr	d6, [pB], #8
+	ldr	d3, [pA, #8]
+	ldr	d7, [pB, #8]
+	ldr	x22, [pA], #16
+	fmul	v16.4s, v0.4s, v4.s[0]
+	ldr	x26, [pB], #16
+	fmul	v17.4s, v1.4s, v4.s[0]
+	ldr	x23, [pA], #8
+	fmul	v18.4s, v0.4s, v4.s[1]
+	ldr	x27, [pB], #8
+	fmul	v19.4s, v1.4s, v4.s[1]
+	fmul	v20.4s, v0.4s, v4.s[2]
+	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
+	fmul	v21.4s, v1.4s, v4.s[2]
+	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
+	fmul	v22.4s, v0.4s, v4.s[3]
+	fmul	v23.4s, v1.4s, v4.s[3]
+	fmul	v24.4s, v0.4s, v5.s[0]
+	fmul	v25.4s, v1.4s, v5.s[0]
+	fmul	v26.4s, v0.4s, v5.s[1]
+	fmul	v27.4s, v1.4s, v5.s[1]
+	fmul	v28.4s, v0.4s, v5.s[2]
+	fmul	v29.4s, v1.4s, v5.s[2]
+	fmul	v30.4s, v0.4s, v5.s[3]
+	fmul	v31.4s, v1.4s, v5.s[3]
+.endm
+
+.macro KERNEL8x8_M1
+	ldr	d2, [pA], #8
+	fmov	v0.d[1], x20
+	ldr	d6, [pB], #8
+	fmov	v4.d[1], x24
+	ldr	d3, [pA, #8]
+	fmov	v1.d[1], x21
+	ldr	d7, [pB, #8]
+	fmov	v5.d[1], x25
+	fmla	v16.4s, v0.4s, v4.s[0]
+	ldr	x22, [pA], #16
+	fmla	v17.4s, v1.4s, v4.s[0]
+	ldr	x26, [pB], #16
+	fmla	v18.4s, v0.4s, v4.s[1]
+	ldr	x23, [pA], #8
+	fmla	v19.4s, v1.4s, v4.s[1]
+	ldr	x27, [pB], #8
+	fmla	v20.4s, v0.4s, v4.s[2]
+	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
+	fmla	v21.4s, v1.4s, v4.s[2]
+	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
+	fmla	v22.4s, v0.4s, v4.s[3]
+	fmla	v23.4s, v1.4s, v4.s[3]
+	fmla	v24.4s, v0.4s, v5.s[0]
+	fmla	v25.4s, v1.4s, v5.s[0]
+	fmla	v26.4s, v0.4s, v5.s[1]
+	fmla	v27.4s, v1.4s, v5.s[1]
+	fmla	v28.4s, v0.4s, v5.s[2]
+	fmla	v29.4s, v1.4s, v5.s[2]
+	fmla	v30.4s, v0.4s, v5.s[3]
+	fmla	v31.4s, v1.4s, v5.s[3]
+.endm
+
+.macro KERNEL8x8_M2
+	ldr	d0, [pA], #8
+	fmov	v2.d[1], x22
+	ldr	d4, [pB], #8
+	fmov	v6.d[1], x26
+	ldr	d1, [pA, #8]
+	fmov	v3.d[1], x23
+	ldr	d5, [pB, #8]
+	fmov	v7.d[1], x27
+	fmla	v16.4s, v2.4s, v6.s[0]
+	ldr	x20, [pA], #16
+	fmla	v17.4s, v3.4s, v6.s[0]
+	ldr	x24, [pB], #16
+	fmla	v18.4s, v2.4s, v6.s[1]
+	ldr	x21, [pA], #8
+	fmla	v19.4s, v3.4s, v6.s[1]
+	ldr	x25, [pB], #8
+	fmla	v20.4s, v2.4s, v6.s[2]
+	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
+	fmla	v21.4s, v3.4s, v6.s[2]
+	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
+	fmla	v22.4s, v2.4s, v6.s[3]
+	fmla	v23.4s, v3.4s, v6.s[3]
+	fmla	v24.4s, v2.4s, v7.s[0]
+	fmla	v25.4s, v3.4s, v7.s[0]
+	fmla	v26.4s, v2.4s, v7.s[1]
+	fmla	v27.4s, v3.4s, v7.s[1]
+	fmla	v28.4s, v2.4s, v7.s[2]
+	fmla	v29.4s, v3.4s, v7.s[2]
+	fmla	v30.4s, v2.4s, v7.s[3]
+	fmla	v31.4s, v3.4s, v7.s[3]
+.endm
+
+.macro KERNEL8x8_E
+	fmov	v2.d[1], x22
+	fmov	v6.d[1], x26
+	fmov	v3.d[1], x23
+	fmov	v7.d[1], x27
+	fmla	v16.4s, v2.4s, v6.s[0]
+	fmla	v17.4s, v3.4s, v6.s[0]
+	fmla	v18.4s, v2.4s, v6.s[1]
+	fmla	v19.4s, v3.4s, v6.s[1]
+	fmla	v20.4s, v2.4s, v6.s[2]
+	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
+	fmla	v21.4s, v3.4s, v6.s[2]
+	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
+	fmla	v22.4s, v2.4s, v6.s[3]
+	fmla	v23.4s, v3.4s, v6.s[3]
+	fmla	v24.4s, v2.4s, v7.s[0]
+	fmla	v25.4s, v3.4s, v7.s[0]
+	fmla	v26.4s, v2.4s, v7.s[1]
+	fmla	v27.4s, v3.4s, v7.s[1]
+	fmla	v28.4s, v2.4s, v7.s[2]
+	fmla	v29.4s, v3.4s, v7.s[2]
+	fmla	v30.4s, v2.4s, v7.s[3]
+	fmla	v31.4s, v3.4s, v7.s[3]
+.endm
+
+.macro KERNEL8x8_SUB
+	ldp	q0, q1, [pA], #32
+	ldp	q4, q5, [pB], #32
+
+	fmla	v16.4s, v0.4s, v4.s[0]
+	fmla	v17.4s, v1.4s, v4.s[0]
+	fmla	v18.4s, v0.4s, v4.s[1]
+	fmla	v19.4s, v1.4s, v4.s[1]
+	fmla	v20.4s, v0.4s, v4.s[2]
+	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
+	fmla	v21.4s, v1.4s, v4.s[2]
+	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
+	fmla	v22.4s, v0.4s, v4.s[3]
+	fmla	v23.4s, v1.4s, v4.s[3]
+	fmla	v24.4s, v0.4s, v5.s[0]
+	fmla	v25.4s, v1.4s, v5.s[0]
+	fmla	v26.4s, v0.4s, v5.s[1]
+	fmla	v27.4s, v1.4s, v5.s[1]
+	fmla	v28.4s, v0.4s, v5.s[2]
+	fmla	v29.4s, v1.4s, v5.s[2]
+	fmla	v30.4s, v0.4s, v5.s[3]
+	fmla	v31.4s, v1.4s, v5.s[3]
+.endm
+
+.macro SAVE8x8
+	prfm	PLDL2KEEP, [pCRow0, #C_PRE_SIZE]
+	add	pCRow1, pCRow0, LDC
+
+	ldp	q0, q1, [pCRow0]
+	fmla	v0.4s, v16.4s, alphaV0
+	fmla	v1.4s, v17.4s, alphaV1
+	stp	q0, q1, [pCRow0]
+
+	prfm	PLDL2KEEP, [pCRow1, #C_PRE_SIZE]
+	add	pCRow2, pCRow1, LDC
+
+	ldp	q2, q3, [pCRow1]
+	fmla	v2.4s, v18.4s, alphaV2
+	fmla	v3.4s, v19.4s, alphaV3
+	stp	q2, q3, [pCRow1]
+
+	prfm	PLDL2KEEP, [pCRow2, #C_PRE_SIZE]
+	add	pCRow1, pCRow2, LDC
+
+	ldp	q4, q5, [pCRow2]
+	fmla	v4.4s, v20.4s, alphaV0
+	fmla	v5.4s, v21.4s, alphaV1
+	stp	q4, q5, [pCRow2]
+
+	prfm	PLDL2KEEP, [pCRow1, #C_PRE_SIZE]
+	add	pCRow2, pCRow1, LDC
+
+	ldp	q6, q7, [pCRow1]
+	fmla	v6.4s, v22.4s, alphaV2
+	fmla	v7.4s, v23.4s, alphaV3
+	stp	q6, q7, [pCRow1]
+
+	prfm	PLDL2KEEP, [pCRow2, #C_PRE_SIZE]
+	add	pCRow1, pCRow2, LDC
+
+	ldp	q0, q1, [pCRow2]
+	fmla	v0.4s, v24.4s, alphaV0
+	fmla	v1.4s, v25.4s, alphaV1
+	stp	q0, q1, [pCRow2]
+
+	prfm	PLDL2KEEP, [pCRow1, #C_PRE_SIZE]
+	add	pCRow2, pCRow1, LDC
+
+	ldp	q2, q3, [pCRow1]
+	fmla	v2.4s, v26.4s, alphaV2
+	fmla	v3.4s, v27.4s, alphaV3
+	stp	q2, q3, [pCRow1]
+
+	prfm	PLDL2KEEP, [pCRow2, #C_PRE_SIZE]
+	add	pCRow1, pCRow2, LDC
+
+	ldp	q4, q5, [pCRow2]
+	fmla	v4.4s, v28.4s, alphaV0
+	fmla	v5.4s, v29.4s, alphaV1
+	stp	q4, q5, [pCRow2]
+
+	prfm	PLDL2KEEP, [pCRow1, #C_PRE_SIZE]
+
+	ldp	q6, q7, [pCRow1]
+	fmla	v6.4s, v30.4s, alphaV2
+	fmla	v7.4s, v31.4s, alphaV3
+	stp	q6, q7, [pCRow1]
+
+	add	pCRow0, pCRow0, #32
+.endm
+
+/******************************************************************************/
+
+.macro INIT4x8
+	fmov		s16, wzr
+	fmov		s18, wzr
+	fmov		s20, wzr
+	fmov		s22, s16
+	fmov		s24, wzr
+	fmov		s26, s16
+	fmov		s28, s18
+	fmov		s30, s20
+.endm
+
+.macro KERNEL4x8_I
+	ldr	q0, [pA], #16
+	ldp	q4, q5, [pB], #32
+
+	ldr	d2, [pA], #8
+	ldr	d6, [pB], #8
+	ldr	d7, [pB, #8]
+	ldr	x22, [pA], #8
+	fmul	v16.4s, v0.4s, v4.s[0]
+	ldr	x26, [pB], #16
+	fmul	v18.4s, v0.4s, v4.s[1]
+	ldr	x27, [pB], #8
+	fmul	v20.4s, v0.4s, v4.s[2]
+	fmul	v22.4s, v0.4s, v4.s[3]
+	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
+	fmul	v24.4s, v0.4s, v5.s[0]
+	fmul	v26.4s, v0.4s, v5.s[1]
+	fmul	v28.4s, v0.4s, v5.s[2]
+	fmul	v30.4s, v0.4s, v5.s[3]
+.endm
+
+.macro KERNEL4x8_M1
+	ldr	d2, [pA], #8
+	fmov	v0.d[1], x20
+	ldr	d6, [pB], #8
+	fmov	v4.d[1], x24
+	ldr	d7, [pB, #8]
+	fmov	v5.d[1], x25
+	ldr	x22, [pA], #8
+	fmla	v16.4s, v0.4s, v4.s[0]
+	ldr	x26, [pB], #16
+	fmla	v18.4s, v0.4s, v4.s[1]
+	ldr	x27, [pB], #8
+	fmla	v20.4s, v0.4s, v4.s[2]
+	fmla	v22.4s, v0.4s, v4.s[3]
+	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
+	fmla	v24.4s, v0.4s, v5.s[0]
+	fmla	v26.4s, v0.4s, v5.s[1]
+	fmla	v28.4s, v0.4s, v5.s[2]
+	fmla	v30.4s, v0.4s, v5.s[3]
+.endm
+
+.macro KERNEL4x8_M2
+	ldr	d0, [pA], #8
+	fmov	v2.d[1], x22
+	ldr	d4, [pB], #8
+	fmov	v6.d[1], x26
+	ldr	d5, [pB, #8]
+	fmov	v7.d[1], x27
+	ldr	x20, [pA], #8
+	fmla	v16.4s, v2.4s, v6.s[0]
+	ldr	x24, [pB], #16
+	fmla	v18.4s, v2.4s, v6.s[1]
+	ldr	x25, [pB], #8
+	fmla	v20.4s, v2.4s, v6.s[2]
+	fmla	v22.4s, v2.4s, v6.s[3]
+	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
+	fmla	v24.4s, v2.4s, v7.s[0]
+	fmla	v26.4s, v2.4s, v7.s[1]
+	fmla	v28.4s, v2.4s, v7.s[2]
+	fmla	v30.4s, v2.4s, v7.s[3]
+.endm
+
+.macro KERNEL4x8_E
+	fmov	v2.d[1], x22
+	fmov	v6.d[1], x26
+	fmov	v7.d[1], x27
+	fmla	v16.4s, v2.4s, v6.s[0]
+	fmla	v18.4s, v2.4s, v6.s[1]
+	fmla	v20.4s, v2.4s, v6.s[2]
+	fmla	v22.4s, v2.4s, v6.s[3]
+	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
+	fmla	v24.4s, v2.4s, v7.s[0]
+	fmla	v26.4s, v2.4s, v7.s[1]
+	fmla	v28.4s, v2.4s, v7.s[2]
+	fmla	v30.4s, v2.4s, v7.s[3]
+.endm
+
+.macro KERNEL4x8_SUB
+	ldr	q0, [pA], #16
+	ldp	q4, q5, [pB], #32
+
+	fmla	v16.4s, v0.4s, v4.s[0]
+	fmla	v18.4s, v0.4s, v4.s[1]
+	fmla	v20.4s, v0.4s, v4.s[2]
+	fmla	v22.4s, v0.4s, v4.s[3]
+	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
+	fmla	v24.4s, v0.4s, v5.s[0]
+	fmla	v26.4s, v0.4s, v5.s[1]
+	fmla	v28.4s, v0.4s, v5.s[2]
+	fmla	v30.4s, v0.4s, v5.s[3]
+.endm
+
+.macro SAVE4x8
+	add	pCRow1, pCRow0, LDC
+
+	ldr	q0, [pCRow0]
+	fmla	v0.4s, v16.4s, alphaV0
+	str	q0, [pCRow0]
+
+	add	pCRow2, pCRow1, LDC
+
+	ldr	q2, [pCRow1]
+	fmla	v2.4s, v18.4s, alphaV2
+	str	q2, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+	ldr	q4, [pCRow2]
+	fmla	v4.4s, v20.4s, alphaV0
+	str	q4, [pCRow2]
+
+	add	pCRow2, pCRow1, LDC
+
+	ldr	q6, [pCRow1]
+	fmla	v6.4s, v22.4s, alphaV2
+	str	q6, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+	ldr	q0, [pCRow2]
+	fmla	v0.4s, v24.4s, alphaV0
+	str	q0, [pCRow2]
+
+	add	pCRow2, pCRow1, LDC
+
+	ldr	q2, [pCRow1]
+	fmla	v2.4s, v26.4s, alphaV2
+	str	q2, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+	ldr	q4, [pCRow2]
+	fmla	v4.4s, v28.4s, alphaV0
+	str	q4, [pCRow2]
+
+	ldr	q6, [pCRow1]
+	fmla	v6.4s, v30.4s, alphaV2
+	str	q6, [pCRow1]
+
+	add	pCRow0, pCRow0, #16
+.endm
+
+/******************************************************************************/
+
+.macro INIT2x8
+	fmov		s16, wzr
+	fmov		s18, wzr
+	fmov		s20, wzr
+	fmov		s22, s16
+	fmov		s24, wzr
+	fmov		s26, s16
+	fmov		s28, s18
+	fmov		s30, s20
+.endm
+
+.macro KERNEL2x8_SUB
+	ldr	d0, [pA], #8
+	ldp	q4, q5, [pB], #32
+
+	fmla	v16.2s, v0.2s, v4.s[0]
+	fmla	v18.2s, v0.2s, v4.s[1]
+	fmla	v20.2s, v0.2s, v4.s[2]
+	fmla	v22.2s, v0.2s, v4.s[3]
+	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
+	fmla	v24.2s, v0.2s, v5.s[0]
+	fmla	v26.2s, v0.2s, v5.s[1]
+	fmla	v28.2s, v0.2s, v5.s[2]
+	fmla	v30.2s, v0.2s, v5.s[3]
+.endm
+
+.macro SAVE2x8
+	add	pCRow1, pCRow0, LDC
+
+	ldr	d0, [pCRow0]
+	fmla	v0.2s, v16.2s, alphaV0
+	str	d0, [pCRow0]
+
+	add	pCRow2, pCRow1, LDC
+
+	ldr	d2, [pCRow1]
+	fmla	v2.2s, v18.2s, alphaV2
+	str	d2, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+	ldr	d4, [pCRow2]
+	fmla	v4.2s, v20.2s, alphaV0
+	str	d4, [pCRow2]
+
+	add	pCRow2, pCRow1, LDC
+
+	ldr	d6, [pCRow1]
+	fmla	v6.2s, v22.2s, alphaV2
+	str	d6, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+	ldr	d0, [pCRow2]
+	fmla	v0.2s, v24.2s, alphaV0
+	str	d0, [pCRow2]
+
+	add	pCRow2, pCRow1, LDC
+
+	ldr	d2, [pCRow1]
+	fmla	v2.2s, v26.2s, alphaV2
+	str	d2, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+	ldr	d4, [pCRow2]
+	fmla	v4.2s, v28.2s, alphaV0
+	str	d4, [pCRow2]
+
+	ldr	d6, [pCRow1]
+	fmla	v6.2s, v30.2s, alphaV2
+	str	d6, [pCRow1]
+
+	add	pCRow0, pCRow0, #8
+.endm
+
+/******************************************************************************/
+
+.macro INIT1x8
+	fmov		s16, wzr
+	fmov		s18, wzr
+	fmov		s20, wzr
+	fmov		s22, s16
+	fmov		s24, wzr
+	fmov		s26, s16
+	fmov		s28, s18
+	fmov		s30, s20
+.endm
+
+.macro KERNEL1x8_SUB
+	ldp	q4, q5, [pB], #32
+	ldr	s0, [pA], #4
+
+	fmla	s16, s0, v4.s[0]
+	fmla	s18, s0, v4.s[1]
+	fmla	s20, s0, v4.s[2]
+	fmla	s22, s0, v4.s[3]
+	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
+	fmla	s24, s0, v5.s[0]
+	fmla	s26, s0, v5.s[1]
+	fmla	s28, s0, v5.s[2]
+	fmla	s30, s0, v5.s[3]
+.endm
+
+.macro SAVE1x8
+	add	pCRow1, pCRow0, LDC
+
+	ldr	s0, [pCRow0]
+	fmla	s0, s16, alphaV0
+	str	s0, [pCRow0]
+
+	add	pCRow2, pCRow1, LDC
+
+	ldr	s2, [pCRow1]
+	fmla	s2, s18, alphaV2
+	str	s2, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+	ldr	s4, [pCRow2]
+	fmla	s4, s20, alphaV0
+	str	s4, [pCRow2]
+
+	add	pCRow2, pCRow1, LDC
+
+	ldr	s6, [pCRow1]
+	fmla	s6, s22, alphaV2
+	str	s6, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+	ldr	s0, [pCRow2]
+	fmla	s0, s24, alphaV0
+	str	s0, [pCRow2]
+
+	add	pCRow2, pCRow1, LDC
+
+	ldr	s2, [pCRow1]
+	fmla	s2, s26, alphaV2
+	str	s2, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+	ldr	s4, [pCRow2]
+	fmla	s4, s28, alphaV0
+	str	s4, [pCRow2]
+
+	ldr	s6, [pCRow1]
+	fmla	s6, s30, alphaV2
+	str	s6, [pCRow1]
+
+	add	pCRow0, pCRow0, #4
+.endm
+
+/******************************************************************************/
+
+.macro INIT8x4
+	fmov		s16, wzr
+	fmov		s17, wzr
+	fmov		s18, wzr
+	fmov		s19, s16
+	fmov		s20, wzr
+	fmov		s21, s16
+	fmov		s22, wzr
+	fmov		s23, s16
+.endm
+
+.macro KERNEL8x4_I
+	ldp	q0, q1, [pA], #32
+	ldr	q4, [pB], #16
+
+	ldr	d2, [pA], #8
+	ldr	d6, [pB], #8
+	ldr	d3, [pA, #8]
+	fmul	v16.4s, v0.4s, v4.s[0]
+	ldr	x22, [pA], #16
+	fmul	v17.4s, v1.4s, v4.s[0]
+	ldr	x26, [pB], #8
+	fmul	v18.4s, v0.4s, v4.s[1]
+	ldr	x23, [pA], #8
+	fmul	v19.4s, v1.4s, v4.s[1]
+	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
+	fmul	v20.4s, v0.4s, v4.s[2]
+	fmul	v21.4s, v1.4s, v4.s[2]
+	fmul	v22.4s, v0.4s, v4.s[3]
+	fmul	v23.4s, v1.4s, v4.s[3]
+.endm
+
+.macro KERNEL8x4_M1
+	ldr	d2, [pA], #8
+	fmov	v0.d[1], x20
+	ldr	d6, [pB], #8
+	fmov	v4.d[1], x24
+	ldr	d3, [pA, #8]
+	fmov	v1.d[1], x21
+	ldr	x22, [pA], #16
+	fmla	v16.4s, v0.4s, v4.s[0]
+	ldr	x26, [pB], #8
+	fmla	v17.4s, v1.4s, v4.s[0]
+	ldr	x23, [pA], #8
+	fmla	v18.4s, v0.4s, v4.s[1]
+	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
+	fmla	v19.4s, v1.4s, v4.s[1]
+	fmla	v20.4s, v0.4s, v4.s[2]
+	fmla	v21.4s, v1.4s, v4.s[2]
+	fmla	v22.4s, v0.4s, v4.s[3]
+	fmla	v23.4s, v1.4s, v4.s[3]
+.endm
+
+.macro KERNEL8x4_M2
+	ldr	d0, [pA], #8
+	fmov	v2.d[1], x22
+	ldr	d4, [pB], #8
+	fmov	v6.d[1], x26
+	ldr	d1, [pA, #8]
+	fmov	v3.d[1], x23
+	ldr	x20, [pA], #16
+	fmla	v16.4s, v2.4s, v6.s[0]
+	ldr	x24, [pB], #8
+	fmla	v17.4s, v3.4s, v6.s[0]
+	ldr	x21, [pA], #8
+	fmla	v18.4s, v2.4s, v6.s[1]
+	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
+	fmla	v19.4s, v3.4s, v6.s[1]
+	fmla	v20.4s, v2.4s, v6.s[2]
+	fmla	v21.4s, v3.4s, v6.s[2]
+	fmla	v22.4s, v2.4s, v6.s[3]
+	fmla	v23.4s, v3.4s, v6.s[3]
+.endm
+
+.macro KERNEL8x4_E
+	fmov	v2.d[1], x22
+	fmov	v6.d[1], x26
+	fmov	v3.d[1], x23
+	fmla	v16.4s, v2.4s, v6.s[0]
+	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
+	fmla	v17.4s, v3.4s, v6.s[0]
+	fmla	v18.4s, v2.4s, v6.s[1]
+	fmla	v19.4s, v3.4s, v6.s[1]
+	fmla	v20.4s, v2.4s, v6.s[2]
+	fmla	v21.4s, v3.4s, v6.s[2]
+	fmla	v22.4s, v2.4s, v6.s[3]
+	fmla	v23.4s, v3.4s, v6.s[3]
+.endm
+
+.macro KERNEL8x4_SUB
+	ldp	q0, q1, [pA], #32
+	ldr	q4, [pB], #16
+
+	fmla	v16.4s, v0.4s, v4.s[0]
+	fmla	v17.4s, v1.4s, v4.s[0]
+	fmla	v18.4s, v0.4s, v4.s[1]
+	fmla	v19.4s, v1.4s, v4.s[1]
+	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
+	fmla	v20.4s, v0.4s, v4.s[2]
+	fmla	v21.4s, v1.4s, v4.s[2]
+	fmla	v22.4s, v0.4s, v4.s[3]
+	fmla	v23.4s, v1.4s, v4.s[3]
+.endm
+
+.macro SAVE8x4
+	prfm	PLDL2KEEP, [pCRow0, #C_PRE_SIZE]
+	add	pCRow1, pCRow0, LDC
+
+	ldp	q0, q1, [pCRow0]
+	fmla	v0.4s, v16.4s, alphaV0
+	fmla	v1.4s, v17.4s, alphaV1
+	stp	q0, q1, [pCRow0]
+
+	prfm	PLDL2KEEP, [pCRow1, #C_PRE_SIZE]
+	add	pCRow2, pCRow1, LDC
+
+	ldp	q4, q5, [pCRow1]
+	fmla	v4.4s, v18.4s, alphaV0
+	fmla	v5.4s, v19.4s, alphaV1
+	stp	q4, q5, [pCRow1]
+
+	prfm	PLDL2KEEP, [pCRow2, #C_PRE_SIZE]
+	add	pCRow1, pCRow2, LDC
+
+	ldp	q0, q1, [pCRow2]
+	fmla	v0.4s, v20.4s, alphaV0
+	fmla	v1.4s, v21.4s, alphaV1
+	stp	q0, q1, [pCRow2]
+
+	prfm	PLDL2KEEP, [pCRow1, #C_PRE_SIZE]
+
+	ldp	q4, q5, [pCRow1]
+	fmla	v4.4s, v22.4s, alphaV0
+	fmla	v5.4s, v23.4s, alphaV1
+	stp	q4, q5, [pCRow1]
+
+	add	pCRow0, pCRow0, #32
+.endm
+
+/******************************************************************************/
+
+
+.macro INIT4x4
+	fmov		s16, wzr
+	fmov		s18, wzr
+	fmov		s20, wzr
+	fmov		s22, wzr
+.endm
+
+.macro KERNEL4x4_I
+	ldr	q0, [pA], #16
+	ldr	q4, [pB], #16
+
+	ldr	d2, [pA], #8
+	ldr	d6, [pB], #8
+	fmul	v16.4s, v0.4s, v4.s[0]
+	ldr	x22, [pA], #8
+	fmul	v18.4s, v0.4s, v4.s[1]
+	ldr	x26, [pB], #8
+	fmul	v20.4s, v0.4s, v4.s[2]
+	fmul	v22.4s, v0.4s, v4.s[3]
+.endm
+
+.macro KERNEL4x4_M1
+	ldr	d2, [pA], #8
+	fmov	v0.d[1], x20
+	ldr	d6, [pB], #8
+	fmov	v4.d[1], x24
+	ldr	x22, [pA], #8
+	ldr	x26, [pB], #8
+	fmla	v16.4s, v0.4s, v4.s[0]
+	fmla	v18.4s, v0.4s, v4.s[1]
+	fmla	v20.4s, v0.4s, v4.s[2]
+	fmla	v22.4s, v0.4s, v4.s[3]
+.endm
+
+.macro KERNEL4x4_M2
+	ldr	d0, [pA], #8
+	fmov	v2.d[1], x22
+	ldr	d4, [pB], #8
+	fmov	v6.d[1], x26
+	ldr	x20, [pA], #8
+	ldr	x24, [pB], #8
+	fmla	v16.4s, v2.4s, v6.s[0]
+	fmla	v18.4s, v2.4s, v6.s[1]
+	fmla	v20.4s, v2.4s, v6.s[2]
+	fmla	v22.4s, v2.4s, v6.s[3]
+.endm
+
+.macro KERNEL4x4_E
+	fmov	v2.d[1], x22
+	fmov	v6.d[1], x26
+	fmla	v16.4s, v2.4s, v6.s[0]
+	fmla	v18.4s, v2.4s, v6.s[1]
+	fmla	v20.4s, v2.4s, v6.s[2]
+	fmla	v22.4s, v2.4s, v6.s[3]
+.endm
+
+.macro KERNEL4x4_SUB
+	ldr	q0, [pA], #16
+	ldr	q4, [pB], #16
+
+	fmla	v16.4s, v0.4s, v4.s[0]
+	fmla	v18.4s, v0.4s, v4.s[1]
+	fmla	v20.4s, v0.4s, v4.s[2]
+	fmla	v22.4s, v0.4s, v4.s[3]
+.endm
+
+.macro SAVE4x4
+	ldr	q0, [pCRow0]
+	fmla	v0.4s, v16.4s, alphaV0
+	str	q0, [pCRow0]
+
+	add	pCRow1, pCRow0, LDC
+	ldr	q1, [pCRow1]
+	fmla	v1.4s, v18.4s, alphaV2
+	str	q1, [pCRow1]
+
+	add	pCRow2, pCRow1, LDC
+	ldr	q2, [pCRow2]
+	fmla	v2.4s, v20.4s, alphaV0
+	str	q2, [pCRow2]
+
+	add	pCRow1, pCRow2, LDC
+	ldr	q3, [pCRow1]
+	fmla	v3.4s, v22.4s, alphaV2
+	str	q3, [pCRow1]
+
+	add	pCRow0, pCRow0, #16
+.endm
+
+/******************************************************************************/
+
+.macro INIT2x4
+	fmov		s16, wzr
+	fmov		s18, wzr
+	fmov		s20, wzr
+	fmov		s22, s16
+.endm
+
+.macro KERNEL2x4_SUB
+	ldr	d0, [pA], #8
+	ldr	q4, [pB], #16
+
+	fmla	v16.2s, v0.2s, v4.s[0]
+	fmla	v18.2s, v0.2s, v4.s[1]
+	fmla	v20.2s, v0.2s, v4.s[2]
+	fmla	v22.2s, v0.2s, v4.s[3]
+.endm
+
+.macro SAVE2x4
+	ldr	d8, [pCRow0]
+	fmla	v8.2s, v16.2s, alphaV0
+	str	d8, [pCRow0]
+
+	add	pCRow1, pCRow0, LDC
+	ldr	d12, [pCRow1]
+	fmla	v12.2s, v18.2s, alphaV1
+	str	d12, [pCRow1]
+
+	add	pCRow2, pCRow1, LDC
+	ldr	d8, [pCRow2]
+	fmla	v8.2s, v20.2s, alphaV2
+	str	d8, [pCRow2]
+
+	add	pCRow1, pCRow2, LDC
+	ldr	d12, [pCRow1]
+	fmla	v12.2s, v22.2s, alphaV3
+	str	d12, [pCRow1]
+
+	add	pCRow0, pCRow0, #8
+.endm
+
+/******************************************************************************/
+
+.macro INIT1x4
+	fmov		s16, wzr
+	fmov		s20, s16
+.endm
+
+.macro KERNEL1x4_SUB
+	ldr	s0, [pA]
+	add	pA, pA, #4
+
+	ld1	{v8.2s, v9.2s}, [pB]
+	add	pB, pB, #16
+
+	fmla	v16.2s, v8.2s, v0.s[0]
+	fmla	v20.2s, v9.2s, v0.s[0]
+.endm
+
+.macro SAVE1x4
+	add	pCRow1, pCRow0, LDC
+	ld1	{v8.s}[0], [pCRow0]
+	ld1	{v8.s}[1], [pCRow1]
+	fmla	v8.2s, v16.2s, alphaV0
+	st1	{v8.s}[0], [pCRow0]
+	st1	{v8.s}[1], [pCRow1]
+
+	add	pCRow2, pCRow1, LDC
+	add	pCRow1, pCRow2, LDC
+	ld1	{v12.s}[0], [pCRow2]
+	ld1	{v12.s}[1], [pCRow1]
+	fmla	v12.2s, v20.2s, alphaV1
+	st1	{v12.s}[0], [pCRow2]
+	st1	{v12.s}[1], [pCRow1]
+
+	add	pCRow0, pCRow0, #4
+.endm
+
+/******************************************************************************/
+
+.macro INIT8x2
+	fmov	s16, wzr
+	fmov	s17, s16
+	fmov	s18, s17
+	fmov	s19, s16
+.endm
+
+.macro KERNEL8x2_SUB
+	ldp	q0, q1, [pA], #32
+	ldr	d4, [pB], #8
+
+	fmla	v16.4s, v0.4s, v4.s[0]
+	fmla	v17.4s, v1.4s, v4.s[0]
+	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
+	fmla	v18.4s, v0.4s, v4.s[1]
+	fmla	v19.4s, v1.4s, v4.s[1]
+.endm
+
+.macro SAVE8x2
+	prfm	PLDL2KEEP, [pCRow0, #C_PRE_SIZE]
+	add	pCRow1, pCRow0, LDC
+
+	ldp	q0, q1, [pCRow0]
+	fmla	v0.4s, v16.4s, alphaV0
+	fmla	v1.4s, v17.4s, alphaV1
+	stp	q0, q1, [pCRow0]
+
+	prfm	PLDL2KEEP, [pCRow1, #C_PRE_SIZE]
+	add	pCRow2, pCRow1, LDC
+
+	ldp	q4, q5, [pCRow1]
+	fmla	v4.4s, v18.4s, alphaV0
+	fmla	v5.4s, v19.4s, alphaV1
+	stp	q4, q5, [pCRow1]
+
+	add	pCRow0, pCRow0, #32
+.endm
+
+/******************************************************************************/
+
+.macro INIT4x2
+	fmov	s16, wzr
+	fmov	s17, s16
+	fmov	s20, s17
+	fmov	s21, s16
+.endm
+
+.macro KERNEL4x2_SUB
+	ld1	{v8.2s}, [pB]
+	add	pB, pB, #8
+	ld1	{v0.2s, v1.2s}, [pA]
+	add	pA, pA, #16
+
+	fmla	v16.2s, v0.2s, v8.s[0]
+	fmla	v17.2s, v1.2s, v8.s[0]
+	fmla	v20.2s, v0.2s, v8.s[1]
+	fmla	v21.2s, v1.2s, v8.s[1]
+.endm
+
+.macro SAVE4x2
+	ld1	{v8.2s, v9.2s}, [pCRow0]
+	fmla	v8.2s, v16.2s, alphaV0
+	fmla	v9.2s, v17.2s, alphaV1
+	st1	{v8.2s, v9.2s}, [pCRow0]
+
+	add	pCRow1, pCRow0, LDC
+	ld1	{v12.2s, v13.2s}, [pCRow1]
+	fmla	v12.2s, v20.2s, alphaV2
+	fmla	v13.2s, v21.2s, alphaV3
+	st1	{v12.2s, v13.2s}, [pCRow1]
+
+	add	pCRow0, pCRow0, #16
+.endm
+
+/******************************************************************************/
+
+.macro INIT2x2
+	fmov		s16, wzr
+	fmov		s20, s16
+.endm
+
+.macro KERNEL2x2_SUB
+	ld1	{v8.2s}, [pB]
+	add	pB, pB, #8
+
+	ld1	{v0.2s}, [pA]
+	add	pA, pA, #8
+
+	fmla	v16.2s, v0.2s, v8.s[0]
+	fmla	v20.2s, v0.2s, v8.s[1]
+.endm
+
+.macro SAVE2x2
+	ld1	{v8.2s}, [pCRow0]
+	fmla	v8.2s, v16.2s, alphaV0
+	st1	{v8.2s}, [pCRow0]
+
+	add	pCRow1 , pCRow0, LDC
+	ld1	{v12.2s}, [pCRow1]
+	fmla	v12.2s, v20.2s, alphaV1
+	st1	{v12.2s}, [pCRow1]
+
+	add	pCRow0, pCRow0, #8
+.endm
+
+/******************************************************************************/
+
+.macro INIT1x2
+	fmov		s16, wzr
+.endm
+
+.macro KERNEL1x2_SUB
+	ld1	{v8.2s} , [pB]
+	add	pB , pB, #8
+
+	ldr	s0 , [pA]
+	add	pA, pA, #4
+
+	fmla	v16.2s, v8.2s, v0.s[0]
+.endm
+
+.macro SAVE1x2
+	add	pCRow1 , pCRow0, LDC
+	ld1	{v8.s}[0], [pCRow0]
+	ld1	{v8.s}[1], [pCRow1]
+	fmla	v8.2s, v16.2s, alphaV0
+	st1	{v8.s}[0], [pCRow0]
+	st1	{v8.s}[1], [pCRow1]
+
+	add	pCRow0, pCRow0, #4
+.endm
+
+/******************************************************************************/
+
+.macro INIT8x1
+	fmov	s16, wzr
+	fmov	s17, wzr
+.endm
+
+.macro KERNEL8x1_SUB
+	ldr	s4, [pB], #4
+	ldp	q0, q1, [pA], #32
+
+	fmla	v16.4s, v0.4s, v4.s[0]
+	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
+	fmla	v17.4s, v1.4s, v4.s[0]
+.endm
+
+.macro SAVE8x1
+	prfm	PLDL2KEEP, [pCRow0, #C_PRE_SIZE]
+
+	ldp	q0, q1, [pCRow0]
+	fmla	v0.4s, v16.4s, alphaV0
+	fmla	v1.4s, v17.4s, alphaV1
+	stp	q0, q1, [pCRow0]
+
+	add	pCRow0, pCRow0, #32
+.endm
+
+/******************************************************************************/
+
+.macro INIT4x1
+	fmov	s16, wzr
+	fmov	s17, s16
+.endm
+
+.macro KERNEL4x1_SUB
+	ldr	s8, [pB]
+	add	pB , pB, #4
+
+	ld1	{v0.2s, v1.2s}, [pA]
+	add	pA , pA, #16
+
+	fmla	v16.2s, v0.2s, v8.s[0]
+	fmla	v17.2s, v1.2s, v8.s[0]
+.endm
+
+.macro SAVE4x1
+	ld1	{v8.2s, v9.2s}, [pCRow0]
+	fmla	v8.2s, v16.2s, alphaV0
+	fmla	v9.2s, v17.2s, alphaV1
+	st1	{v8.2s, v9.2s}, [pCRow0]
+
+	add	pCRow0, pCRow0, #16
+.endm
+
+/******************************************************************************/
+
+.macro INIT2x1
+	fmov		s16, wzr
+.endm
+
+.macro KERNEL2x1_SUB
+	ldr	s8, [pB]
+	add	pB , pB, #4
+
+	ld1	{v0.2s}, [pA]
+	add	pA , pA, #8
+
+	fmla	v16.2s, v0.2s, v8.s[0]
+.endm
+
+.macro SAVE2x1
+	ld1	{v8.2s}, [pCRow0]
+	fmla	v8.2s, v16.2s, alphaV0
+	st1	{v8.2s}, [pCRow0]
+
+	add	pCRow0, pCRow0, #8
+.endm
+
+/******************************************************************************/
+
+.macro INIT1x1
+	fmov	s16, wzr
+.endm
+
+.macro KERNEL1x1_SUB
+	ldr	s8, [pB]
+	add	pB , pB, #4
+
+	ldr	s0, [pA]
+	add	pA , pA, #4
+
+	fmadd	s16, s0, s8, s16
+.endm
+
+.macro SAVE1x1
+	ldr	s8, [pCRow0]
+	fmla	s8, s16, alphaV0
+	str	s8, [pCRow0]
+
+	add	pCRow0, pCRow0, #4
+.endm
+
+/*******************************************************************************
+* End of macro definitions
+*******************************************************************************/
+
+	PROLOGUE
+
+.Lsgemm_kernel_begin:
+
+	.align 5
+	add	sp, sp, #-(11 * 16)
+	stp	d8, d9, [sp, #(0 * 16)]
+	stp	d10, d11, [sp, #(1 * 16)]
+	stp	d12, d13, [sp, #(2 * 16)]
+	stp	d14, d15, [sp, #(3 * 16)]
+	stp	d16, d17, [sp, #(4 * 16)]
+	stp	x18, x19, [sp, #(5 * 16)]
+	stp	x20, x21, [sp, #(6 * 16)]
+	stp	x22, x23, [sp, #(7 * 16)]
+	stp	x24, x25, [sp, #(8 * 16)]
+	stp	x26, x27, [sp, #(9 * 16)]
+	str	x28, [sp, #(10 * 16)]
+
+	fmov	alpha0, s0
+	fmov	alpha1, s0
+	fmov	alpha2, s0
+	fmov	alpha3, s0
+
+	lsl	LDC, LDC, #2			// ldc = ldc * 4
+
+	mov	pB, origPB
+
+	mov	counterJ, origN
+	asr	counterJ, counterJ, #3		// J = J / 8
+	cmp	counterJ, #0
+	ble	.Lsgemm_kernel_L4_BEGIN
+
+/******************************************************************************/
+/******************************************************************************/
+
+.Lsgemm_kernel_L8_BEGIN:
+	mov	pCRow0, pC			// pCRow0 = C
+	add	pC, pC, LDC, lsl #3
+
+	mov	pA, origPA			// pA = start of A array
+
+/******************************************************************************/
+
+.Lsgemm_kernel_L8_M8_BEGIN:
+
+	mov	counterI, origM
+	asr	counterI, counterI, #3		// counterI = counterI / 8
+	cmp	counterI, #0
+	ble	.Lsgemm_kernel_L8_M4_BEGIN
+
+.Lsgemm_kernel_L8_M8_20:
+
+	mov	pB, origPB
+
+	asr	counterL , origK, #3		// L = K / 8
+	cmp	counterL , #2			// is there at least 16 to do?
+	blt	.Lsgemm_kernel_L8_M8_32
+
+	KERNEL8x8_I				// do one in the K
+	KERNEL8x8_M2				// do another in the K
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+
+	subs	counterL, counterL, #2
+	ble	.Lsgemm_kernel_L8_M8_22a
+	.align 5
+
+.Lsgemm_kernel_L8_M8_22:
+
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L8_M8_22
+
+.Lsgemm_kernel_L8_M8_22a:
+
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_E
+
+	b	 .Lsgemm_kernel_L8_M8_44
+
+.Lsgemm_kernel_L8_M8_32:
+
+	tst	counterL, #1
+	ble	.Lsgemm_kernel_L8_M8_40
+
+	KERNEL8x8_I
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_E
+
+	b	.Lsgemm_kernel_L8_M8_44
+
+.Lsgemm_kernel_L8_M8_40:
+
+	INIT8x8
+
+.Lsgemm_kernel_L8_M8_44:
+
+	ands	counterL , origK, #7
+	ble	.Lsgemm_kernel_L8_M8_100
+
+.Lsgemm_kernel_L8_M8_46:
+
+	KERNEL8x8_SUB
+
+	subs	counterL, counterL, 1
+	bgt	.Lsgemm_kernel_L8_M8_46
+
+.Lsgemm_kernel_L8_M8_100:
+
+	SAVE8x8
+
+.Lsgemm_kernel_L8_M8_END:
+	subs	counterI, counterI, #1
+	bne	.Lsgemm_kernel_L8_M8_20
+
+/******************************************************************************/
+
+.Lsgemm_kernel_L8_M4_BEGIN:
+
+	mov	counterI, origM
+	tst	counterI , #7
+	ble	.Lsgemm_kernel_L8_END
+
+	tst	counterI, #4
+	ble	.Lsgemm_kernel_L8_M2_BEGIN
+
+.Lsgemm_kernel_L8_M4_20:
+
+	mov	pB, origPB
+
+	asr	counterL , origK, #1		// L = K / 2
+	cmp	counterL , #2			// is there at least 4 to do?
+	blt	.Lsgemm_kernel_L8_M4_32
+
+	KERNEL4x8_I				// do one in the K
+	KERNEL4x8_M2				// do another in the K
+
+	subs	counterL, counterL, #2
+	ble	.Lsgemm_kernel_L8_M4_22a
+	.align 5
+
+.Lsgemm_kernel_L8_M4_22:
+
+	KERNEL4x8_M1
+	KERNEL4x8_M2
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L8_M4_22
+
+.Lsgemm_kernel_L8_M4_22a:
+
+	KERNEL4x8_M1
+	KERNEL4x8_E
+
+	b	 .Lsgemm_kernel_L8_M4_44
+
+.Lsgemm_kernel_L8_M4_32:
+
+	tst	counterL, #1
+	ble	.Lsgemm_kernel_L8_M4_40
+
+	KERNEL4x8_I
+	KERNEL4x8_E
+
+	b	.Lsgemm_kernel_L8_M4_44
+
+.Lsgemm_kernel_L8_M4_40:
+
+	INIT4x8
+
+.Lsgemm_kernel_L8_M4_44:
+
+	ands	counterL , origK, #1
+	ble	.Lsgemm_kernel_L8_M4_100
+
+.Lsgemm_kernel_L8_M4_46:
+
+	KERNEL4x8_SUB
+
+.Lsgemm_kernel_L8_M4_100:
+
+	SAVE4x8
+
+.Lsgemm_kernel_L8_M4_END:
+
+/******************************************************************************/
+
+.Lsgemm_kernel_L8_M2_BEGIN:
+
+	mov	counterI, origM
+	tst	counterI , #3
+	ble	.Lsgemm_kernel_L8_END
+
+	tst	counterI, #2			// counterI = counterI / 2
+	ble	.Lsgemm_kernel_L8_M1_BEGIN
+
+.Lsgemm_kernel_L8_M2_20:
+
+	INIT2x8
+
+	mov	pB, origPB
+
+	asr	counterL , origK, #3		// counterL = counterL / 8
+	cmp	counterL , #0
+	ble	.Lsgemm_kernel_L8_M2_40
+
+.Lsgemm_kernel_L8_M2_22:
+
+	KERNEL2x8_SUB
+	KERNEL2x8_SUB
+	KERNEL2x8_SUB
+	KERNEL2x8_SUB
+
+	KERNEL2x8_SUB
+	KERNEL2x8_SUB
+	KERNEL2x8_SUB
+	KERNEL2x8_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L8_M2_22
+
+
+.Lsgemm_kernel_L8_M2_40:
+
+	ands	counterL , origK, #7		// counterL = counterL % 8
+	ble	.Lsgemm_kernel_L8_M2_100
+
+.Lsgemm_kernel_L8_M2_42:
+
+	KERNEL2x8_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L8_M2_42
+
+.Lsgemm_kernel_L8_M2_100:
+
+	SAVE2x8
+
+.Lsgemm_kernel_L8_M2_END:
+
+/******************************************************************************/
+
+.Lsgemm_kernel_L8_M1_BEGIN:
+
+	tst	counterI, #1			// counterI = counterI % 2
+	ble	.Lsgemm_kernel_L8_END
+
+.Lsgemm_kernel_L8_M1_20:
+
+	INIT1x8
+
+	mov	pB, origPB
+
+	asr	counterL , origK, #3		// counterL = counterL / 8
+	cmp	counterL , #0
+	ble	.Lsgemm_kernel_L8_M1_40
+
+.Lsgemm_kernel_L8_M1_22:
+	KERNEL1x8_SUB
+	KERNEL1x8_SUB
+	KERNEL1x8_SUB
+	KERNEL1x8_SUB
+
+	KERNEL1x8_SUB
+	KERNEL1x8_SUB
+	KERNEL1x8_SUB
+	KERNEL1x8_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L8_M1_22
+
+
+.Lsgemm_kernel_L8_M1_40:
+
+	ands	counterL , origK, #7		// counterL = counterL % 8
+	ble	.Lsgemm_kernel_L8_M1_100
+
+.Lsgemm_kernel_L8_M1_42:
+
+	KERNEL1x8_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L8_M1_42
+
+.Lsgemm_kernel_L8_M1_100:
+
+	SAVE1x8
+
+.Lsgemm_kernel_L8_END:
+	lsl	temp, origK, #5			// B = B + K * 4 * 8
+	add	origPB, origPB, temp
+
+	subs	counterJ, counterJ , #1		// j--
+	bgt	.Lsgemm_kernel_L8_BEGIN
+
+/******************************************************************************/
+/******************************************************************************/
+
+.Lsgemm_kernel_L4_BEGIN:
+
+	mov	counterJ , origN
+	tst	counterJ , #7
+	ble	.Lsgemm_kernel_L999
+
+	tst	counterJ , #4
+	ble	.Lsgemm_kernel_L2_BEGIN
+
+	mov	pCRow0, pC			// pCRow0 = pC
+
+	add	pC,pC,LDC, lsl #2
+
+	mov	pA, origPA			// pA = A
+
+/******************************************************************************/
+
+.Lsgemm_kernel_L4_M8_BEGIN:
+
+	mov	counterI, origM
+	asr	counterI, counterI, #3		// counterI = counterI / 8
+	cmp	counterI, #0
+	ble	.Lsgemm_kernel_L4_M4_BEGIN
+
+.Lsgemm_kernel_L4_M8_20:
+
+	mov	pB, origPB
+
+	asr	counterL , origK, #1		// L = K / 2
+	cmp	counterL , #2			// is there at least 4 to do?
+	blt	.Lsgemm_kernel_L4_M8_32
+
+	KERNEL8x4_I				// do one in the K
+	KERNEL8x4_M2				// do another in the K
+
+	subs	counterL, counterL, #2
+	ble	.Lsgemm_kernel_L4_M8_22a
+	.align 5
+
+.Lsgemm_kernel_L4_M8_22:
+
+	KERNEL8x4_M1
+	KERNEL8x4_M2
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L4_M8_22
+
+.Lsgemm_kernel_L4_M8_22a:
+
+	KERNEL8x4_M1
+	KERNEL8x4_E
+
+	b	 .Lsgemm_kernel_L4_M8_44
+
+.Lsgemm_kernel_L4_M8_32:
+
+	tst	counterL, #1
+	ble	.Lsgemm_kernel_L4_M8_40
+
+	KERNEL8x4_I
+	KERNEL8x4_E
+
+	b	.Lsgemm_kernel_L4_M8_44
+
+.Lsgemm_kernel_L4_M8_40:
+
+	INIT8x4
+
+.Lsgemm_kernel_L4_M8_44:
+
+	ands	counterL , origK, #1
+	ble	.Lsgemm_kernel_L4_M8_100
+
+.Lsgemm_kernel_L4_M8_46:
+
+	KERNEL8x4_SUB
+
+.Lsgemm_kernel_L4_M8_100:
+
+	SAVE8x4
+
+.Lsgemm_kernel_L4_M8_END:
+	subs	counterI, counterI, #1
+	bne	.Lsgemm_kernel_L4_M8_20
+
+/******************************************************************************/
+
+.Lsgemm_kernel_L4_M4_BEGIN:
+
+	mov	counterI, origM
+	tst	counterI , #7
+	ble	.Lsgemm_kernel_L4_END
+
+	tst	counterI, #4
+	ble	.Lsgemm_kernel_L4_M2_BEGIN
+
+.Lsgemm_kernel_L4_M4_20:
+
+	mov	pB, origPB
+
+	asr	counterL , origK, #1		// L = K / 2
+	cmp	counterL , #2			// is there at least 4 to do?
+	blt	.Lsgemm_kernel_L4_M4_32
+
+	KERNEL4x4_I				// do one in the K
+	KERNEL4x4_M2				// do another in the K
+
+	subs	counterL, counterL, #2
+	ble	.Lsgemm_kernel_L4_M4_22a
+	.align 5
+
+.Lsgemm_kernel_L4_M4_22:
+
+	KERNEL4x4_M1
+	KERNEL4x4_M2
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L4_M4_22
+
+.Lsgemm_kernel_L4_M4_22a:
+
+	KERNEL4x4_M1
+	KERNEL4x4_E
+
+	b	 .Lsgemm_kernel_L4_M4_44
+
+.Lsgemm_kernel_L4_M4_32:
+
+	tst	counterL, #1
+	ble	.Lsgemm_kernel_L4_M4_40
+
+	KERNEL4x4_I
+	KERNEL4x4_E
+
+	b	.Lsgemm_kernel_L4_M4_44
+
+.Lsgemm_kernel_L4_M4_40:
+
+	INIT4x4
+
+.Lsgemm_kernel_L4_M4_44:
+
+	ands	counterL , origK, #1
+	ble	.Lsgemm_kernel_L4_M4_100
+
+.Lsgemm_kernel_L4_M4_46:
+
+	KERNEL4x4_SUB
+
+.Lsgemm_kernel_L4_M4_100:
+
+	SAVE4x4
+
+.Lsgemm_kernel_L4_M4_END:
+
+/******************************************************************************/
+
+.Lsgemm_kernel_L4_M2_BEGIN:
+
+	mov	counterI, origM
+	tst	counterI , #3
+	ble	.Lsgemm_kernel_L4_END
+
+	tst	counterI, #2			// counterI = counterI / 2
+	ble	.Lsgemm_kernel_L4_M1_BEGIN
+
+.Lsgemm_kernel_L4_M2_20:
+
+	INIT2x4
+
+	mov	pB, origPB
+
+	asr	counterL , origK, #3		// counterL = counterL / 8
+	cmp	counterL , #0
+	ble	.Lsgemm_kernel_L4_M2_40
+
+.Lsgemm_kernel_L4_M2_22:
+
+	KERNEL2x4_SUB
+	KERNEL2x4_SUB
+	KERNEL2x4_SUB
+	KERNEL2x4_SUB
+
+	KERNEL2x4_SUB
+	KERNEL2x4_SUB
+	KERNEL2x4_SUB
+	KERNEL2x4_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L4_M2_22
+
+
+.Lsgemm_kernel_L4_M2_40:
+
+	ands	counterL , origK, #7		// counterL = counterL % 8
+	ble	.Lsgemm_kernel_L4_M2_100
+
+.Lsgemm_kernel_L4_M2_42:
+
+	KERNEL2x4_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L4_M2_42
+
+.Lsgemm_kernel_L4_M2_100:
+
+	SAVE2x4
+
+.Lsgemm_kernel_L4_M2_END:
+
+/******************************************************************************/
+
+.Lsgemm_kernel_L4_M1_BEGIN:
+
+	tst	counterI, #1			// counterI = counterI % 2
+	ble	.Lsgemm_kernel_L4_END
+
+.Lsgemm_kernel_L4_M1_20:
+
+	INIT1x4
+
+	mov	pB, origPB
+
+	asr	counterL , origK, #3		// counterL = counterL / 8
+	cmp	counterL , #0
+	ble	.Lsgemm_kernel_L4_M1_40
+
+.Lsgemm_kernel_L4_M1_22:
+	KERNEL1x4_SUB
+	KERNEL1x4_SUB
+	KERNEL1x4_SUB
+	KERNEL1x4_SUB
+
+	KERNEL1x4_SUB
+	KERNEL1x4_SUB
+	KERNEL1x4_SUB
+	KERNEL1x4_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L4_M1_22
+
+
+.Lsgemm_kernel_L4_M1_40:
+
+	ands	counterL , origK, #7		// counterL = counterL % 8
+	ble	.Lsgemm_kernel_L4_M1_100
+
+.Lsgemm_kernel_L4_M1_42:
+
+	KERNEL1x4_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L4_M1_42
+
+.Lsgemm_kernel_L4_M1_100:
+
+	SAVE1x4
+
+.Lsgemm_kernel_L4_END:
+	add	origPB, origPB, origK, lsl #4	// B = B + K * 4 * 4
+
+/******************************************************************************/
+/******************************************************************************/
+
+.Lsgemm_kernel_L2_BEGIN:   // less than 2 left in N direction
+
+	mov	counterJ , origN
+	tst	counterJ , #3
+	ble	.Lsgemm_kernel_L999
+
+	tst	counterJ , #2
+	ble	.Lsgemm_kernel_L1_BEGIN
+
+	mov	pCRow0, pC			// pCRow0 = pC
+
+	add	pC,pC,LDC, lsl #1
+
+	mov	pA, origPA			// pA = A
+
+/******************************************************************************/
+
+.Lsgemm_kernel_L2_M8_BEGIN:
+
+	mov	counterI, origM
+	asr	counterI, counterI, #3		// counterI = counterI / 8
+	cmp	counterI,#0
+	ble	.Lsgemm_kernel_L2_M4_BEGIN
+
+.Lsgemm_kernel_L2_M8_20:
+
+	INIT8x2
+
+	mov	pB, origPB
+
+	asr	counterL , origK, #3		// counterL = counterL / 8
+	cmp	counterL,#0
+	ble	.Lsgemm_kernel_L2_M8_40
+	.align 5
+
+.Lsgemm_kernel_L2_M8_22:
+	KERNEL8x2_SUB
+	KERNEL8x2_SUB
+	KERNEL8x2_SUB
+	KERNEL8x2_SUB
+
+	KERNEL8x2_SUB
+	KERNEL8x2_SUB
+	KERNEL8x2_SUB
+	KERNEL8x2_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L2_M8_22
+
+
+.Lsgemm_kernel_L2_M8_40:
+
+	ands	counterL , origK, #7		// counterL = counterL % 8
+	ble	.Lsgemm_kernel_L2_M8_100
+
+.Lsgemm_kernel_L2_M8_42:
+
+	KERNEL8x2_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L2_M8_42
+
+.Lsgemm_kernel_L2_M8_100:
+
+	SAVE8x2
+
+.Lsgemm_kernel_L2_M8_END:
+
+	subs	counterI, counterI, #1
+	bgt	.Lsgemm_kernel_L2_M8_20
+
+/******************************************************************************/
+
+.Lsgemm_kernel_L2_M4_BEGIN:
+
+	mov	counterI, origM
+	tst	counterI , #7
+	ble	.Lsgemm_kernel_L2_END
+
+	tst	counterI, #4
+	ble	.Lsgemm_kernel_L2_M2_BEGIN
+
+.Lsgemm_kernel_L2_M4_20:
+
+	INIT4x2
+
+	mov	pB, origPB
+
+	asr	counterL , origK, #3		// counterL = counterL / 8
+	cmp	counterL,#0
+	ble	.Lsgemm_kernel_L2_M4_40
+	.align 5
+
+.Lsgemm_kernel_L2_M4_22:
+	KERNEL4x2_SUB
+	KERNEL4x2_SUB
+	KERNEL4x2_SUB
+	KERNEL4x2_SUB
+
+	KERNEL4x2_SUB
+	KERNEL4x2_SUB
+	KERNEL4x2_SUB
+	KERNEL4x2_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L2_M4_22
+
+
+.Lsgemm_kernel_L2_M4_40:
+
+	ands	counterL , origK, #7		// counterL = counterL % 8
+	ble	.Lsgemm_kernel_L2_M4_100
+
+.Lsgemm_kernel_L2_M4_42:
+
+	KERNEL4x2_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L2_M4_42
+
+.Lsgemm_kernel_L2_M4_100:
+
+	SAVE4x2
+
+.Lsgemm_kernel_L2_M4_END:
+
+/******************************************************************************/
+
+.Lsgemm_kernel_L2_M2_BEGIN:
+
+	mov	counterI, origM
+	tst	counterI , #3
+	ble	.Lsgemm_kernel_L2_END
+
+	tst	counterI, #2			// counterI = counterI / 2
+	ble	.Lsgemm_kernel_L2_M1_BEGIN
+
+.Lsgemm_kernel_L2_M2_20:
+
+	INIT2x2
+
+	mov	pB, origPB
+
+	asr	counterL , origK, #3		// counterL = counterL / 8
+        cmp	counterL,#0
+	ble	.Lsgemm_kernel_L2_M2_40
+
+.Lsgemm_kernel_L2_M2_22:
+
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L2_M2_22
+
+
+.Lsgemm_kernel_L2_M2_40:
+
+	ands	counterL , origK, #7		// counterL = counterL % 8
+	ble	.Lsgemm_kernel_L2_M2_100
+
+.Lsgemm_kernel_L2_M2_42:
+
+	KERNEL2x2_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L2_M2_42
+
+.Lsgemm_kernel_L2_M2_100:
+
+	SAVE2x2
+
+.Lsgemm_kernel_L2_M2_END:
+
+/******************************************************************************/
+
+.Lsgemm_kernel_L2_M1_BEGIN:
+
+	tst	counterI, #1			// counterI = counterI % 2
+	ble	.Lsgemm_kernel_L2_END
+
+.Lsgemm_kernel_L2_M1_20:
+
+	INIT1x2
+
+	mov	pB, origPB
+
+	asr	counterL , origK, #3		// counterL = counterL / 8
+        cmp     counterL, #0
+	ble	.Lsgemm_kernel_L2_M1_40
+
+.Lsgemm_kernel_L2_M1_22:
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L2_M1_22
+
+
+.Lsgemm_kernel_L2_M1_40:
+
+	ands	counterL , origK, #7		// counterL = counterL % 8
+	ble	.Lsgemm_kernel_L2_M1_100
+
+.Lsgemm_kernel_L2_M1_42:
+
+	KERNEL1x2_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L2_M1_42
+
+.Lsgemm_kernel_L2_M1_100:
+
+	SAVE1x2
+
+.Lsgemm_kernel_L2_END:
+
+	add	origPB, origPB, origK, lsl #3	// B = B + K * 2 * 4
+
+/******************************************************************************/
+/******************************************************************************/
+
+.Lsgemm_kernel_L1_BEGIN:
+
+	mov	counterJ , origN
+	tst	counterJ , #1
+	ble	.Lsgemm_kernel_L999 // done
+
+
+	mov	pCRow0, pC			// pCRow0 = C
+	add	pC , pC , LDC			// Update pC to point to next
+
+	mov	pA, origPA			// pA = A
+
+/******************************************************************************/
+
+.Lsgemm_kernel_L1_M8_BEGIN:
+
+	mov	counterI, origM
+	asr	counterI, counterI, #3
+	cmp	counterI, #0
+	ble	.Lsgemm_kernel_L1_M4_BEGIN
+
+.Lsgemm_kernel_L1_M8_20:
+
+	INIT8x1
+
+	mov	pB, origPB
+
+	asr	counterL , origK, #3		// counterL = counterL / 8
+	cmp	counterL , #0
+	ble	.Lsgemm_kernel_L1_M8_40
+	.align 5
+
+.Lsgemm_kernel_L1_M8_22:
+	KERNEL8x1_SUB
+	KERNEL8x1_SUB
+	KERNEL8x1_SUB
+	KERNEL8x1_SUB
+
+	KERNEL8x1_SUB
+	KERNEL8x1_SUB
+	KERNEL8x1_SUB
+	KERNEL8x1_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L1_M8_22
+
+
+.Lsgemm_kernel_L1_M8_40:
+
+	ands	counterL , origK, #7		// counterL = counterL % 8
+	ble	.Lsgemm_kernel_L1_M8_100
+
+.Lsgemm_kernel_L1_M8_42:
+
+	KERNEL8x1_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L1_M8_42
+
+.Lsgemm_kernel_L1_M8_100:
+
+	SAVE8x1
+
+.Lsgemm_kernel_L1_M8_END:
+
+	subs	counterI, counterI, #1
+	bgt	.Lsgemm_kernel_L1_M8_20
+
+/******************************************************************************/
+
+.Lsgemm_kernel_L1_M4_BEGIN:
+
+	mov	counterI, origM
+	tst	counterI , #7
+	ble	.Lsgemm_kernel_L1_END
+
+	tst	counterI, #4
+	ble	.Lsgemm_kernel_L1_M2_BEGIN
+
+.Lsgemm_kernel_L1_M4_20:
+
+	INIT4x1
+
+	mov	pB, origPB
+
+	asr	counterL , origK, #3		// counterL = counterL / 8
+	cmp	counterL , #0
+	ble	.Lsgemm_kernel_L1_M4_40
+	.align 5
+
+.Lsgemm_kernel_L1_M4_22:
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L1_M4_22
+
+
+.Lsgemm_kernel_L1_M4_40:
+
+	ands	counterL , origK, #7		// counterL = counterL % 8
+	ble	.Lsgemm_kernel_L1_M4_100
+
+.Lsgemm_kernel_L1_M4_42:
+
+	KERNEL4x1_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L1_M4_42
+
+.Lsgemm_kernel_L1_M4_100:
+
+	SAVE4x1
+
+.Lsgemm_kernel_L1_M4_END:
+
+/******************************************************************************/
+
+.Lsgemm_kernel_L1_M2_BEGIN:
+
+	mov	counterI, origM
+	tst	counterI , #3
+	ble	.Lsgemm_kernel_L1_END
+
+	tst	counterI, #2			// counterI = counterI / 2
+	ble	.Lsgemm_kernel_L1_M1_BEGIN
+
+.Lsgemm_kernel_L1_M2_20:
+
+	INIT2x1
+
+	mov	pB, origPB
+
+	asr	counterL , origK, #3		// counterL = counterL / 8
+	cmp	counterL , #0
+	ble	.Lsgemm_kernel_L1_M2_40
+
+.Lsgemm_kernel_L1_M2_22:
+
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L1_M2_22
+
+
+.Lsgemm_kernel_L1_M2_40:
+
+	ands	counterL , origK, #7		// counterL = counterL % 8
+	ble	.Lsgemm_kernel_L1_M2_100
+
+.Lsgemm_kernel_L1_M2_42:
+
+	KERNEL2x1_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L1_M2_42
+
+.Lsgemm_kernel_L1_M2_100:
+
+	SAVE2x1
+
+.Lsgemm_kernel_L1_M2_END:
+
+/******************************************************************************/
+
+.Lsgemm_kernel_L1_M1_BEGIN:
+
+	tst	counterI, #1			// counterI = counterI % 2
+	ble	.Lsgemm_kernel_L1_END
+
+.Lsgemm_kernel_L1_M1_20:
+
+	INIT1x1
+
+	mov	pB, origPB
+
+	asr	counterL , origK, #3		// counterL = counterL / 8
+	cmp	counterL , #0
+	ble	.Lsgemm_kernel_L1_M1_40
+
+.Lsgemm_kernel_L1_M1_22:
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L1_M1_22
+
+
+.Lsgemm_kernel_L1_M1_40:
+
+	ands	counterL , origK, #7		// counterL = counterL % 8
+	ble	.Lsgemm_kernel_L1_M1_100
+
+.Lsgemm_kernel_L1_M1_42:
+
+	KERNEL1x1_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lsgemm_kernel_L1_M1_42
+
+.Lsgemm_kernel_L1_M1_100:
+
+	SAVE1x1
+
+.Lsgemm_kernel_L1_END:
+
+/******************************************************************************/
+
+.Lsgemm_kernel_L999:
+	mov	x0, #0				// set return value
+	ldp	d8, d9, [sp, #(0 * 16)]
+	ldp	d10, d11, [sp, #(1 * 16)]
+	ldp	d12, d13, [sp, #(2 * 16)]
+	ldp	d14, d15, [sp, #(3 * 16)]
+	ldp	d16, d17, [sp, #(4 * 16)]
+	ldp	x18, x19, [sp, #(5 * 16)]
+	ldp	x20, x21, [sp, #(6 * 16)]
+	ldp	x22, x23, [sp, #(7 * 16)]
+	ldp	x24, x25, [sp, #(8 * 16)]
+	ldp	x26, x27, [sp, #(9 * 16)]
+	ldr	x28, [sp, #(10 * 16)]
+	add	sp, sp, #(11*16)
+	ret
+
+	EPILOGUE
+
diff --git a/kernel/arm64/sgemm_ncopy_8.S b/kernel/arm64/sgemm_ncopy_8.S
new file mode 100644
index 000000000..f99b1d992
--- /dev/null
+++ b/kernel/arm64/sgemm_ncopy_8.S
@@ -0,0 +1,562 @@
+/***************************************************************************
+Copyright (c) 2016, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A00 PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+#define ASSEMBLER
+#include "common.h"
+
+#define	M	x0
+#define	N	x1
+#define	A00	x2
+#define	LDA	x3
+#define	B00	x4
+
+#define	A01	x5
+#define	A02	x6
+#define	A03	x7
+#define	A04	x8
+#define	A05	x9
+#define	A06	x10
+#define	A07	x11
+#define	A08	x12
+
+#define I	x13
+#define	J	x14
+#define	K	x15
+
+#define	TEMP1	x16
+#define	TEMP2	x17
+
+/**************************************************************************************
+* Macro definitions
+**************************************************************************************/
+
+.macro SAVE_REGS
+	add	sp, sp, #-(11 * 16)
+	stp	d8, d9, [sp, #(0 * 16)]
+	stp	d10, d11, [sp, #(1 * 16)]
+	stp	d12, d13, [sp, #(2 * 16)]
+	stp	d14, d15, [sp, #(3 * 16)]
+	stp	d16, d17, [sp, #(4 * 16)]
+	stp	x18, x19, [sp, #(5 * 16)]
+	stp	x20, x21, [sp, #(6 * 16)]
+	stp	x22, x23, [sp, #(7 * 16)]
+	stp	x24, x25, [sp, #(8 * 16)]
+	stp	x26, x27, [sp, #(9 * 16)]
+	str	x28, [sp, #(10 * 16)]
+.endm
+
+.macro RESTORE_REGS
+	ldp	d8, d9, [sp, #(0 * 16)]
+	ldp	d10, d11, [sp, #(1 * 16)]
+	ldp	d12, d13, [sp, #(2 * 16)]
+	ldp	d14, d15, [sp, #(3 * 16)]
+	ldp	d16, d17, [sp, #(4 * 16)]
+	ldp	x18, x19, [sp, #(5 * 16)]
+	ldp	x20, x21, [sp, #(6 * 16)]
+	ldp	x22, x23, [sp, #(7 * 16)]
+	ldp	x24, x25, [sp, #(8 * 16)]
+	ldp	x26, x27, [sp, #(9 * 16)]
+	ldr	x28, [sp, #(10 * 16)]
+	add	sp, sp, #(11*16)
+.endm
+
+.macro COPY4x8
+	ldr	q0, [A01], #16
+	ldr	q1, [A02], #16
+	ins	v8.s[0], v0.s[0]
+	ins	v10.s[0], v0.s[1]
+	ins	v12.s[0], v0.s[2]
+	ins	v14.s[0], v0.s[3]
+	ins	v8.s[1], v1.s[0]
+	ins	v10.s[1], v1.s[1]
+	ins	v12.s[1], v1.s[2]
+	ins	v14.s[1], v1.s[3]
+
+	ldr	q2, [A03], #16
+	ldr	q3, [A04], #16
+	ins	v8.s[2], v2.s[0]
+	ins	v10.s[2], v2.s[1]
+	ins	v12.s[2], v2.s[2]
+	ins	v14.s[2], v2.s[3]
+	ins	v8.s[3], v3.s[0]
+	ins	v10.s[3], v3.s[1]
+	ins	v12.s[3], v3.s[2]
+	ins	v14.s[3], v3.s[3]
+
+	ldr	q4, [A05], #16
+	ldr	q5, [A06], #16
+	ins	v9.s[0], v4.s[0]
+	ins	v11.s[0], v4.s[1]
+	ins	v13.s[0], v4.s[2]
+	ins	v15.s[0], v4.s[3]
+	ins	v9.s[1], v5.s[0]
+	ins	v11.s[1], v5.s[1]
+	ins	v13.s[1], v5.s[2]
+	ins	v15.s[1], v5.s[3]
+
+	ldr	q6, [A07], #16
+	ldr	q7, [A08], #16
+	ins	v9.s[2], v6.s[0]
+	ins	v11.s[2], v6.s[1]
+	ins	v13.s[2], v6.s[2]
+	ins	v15.s[2], v6.s[3]
+	ins	v9.s[3], v7.s[0]
+	ins	v11.s[3], v7.s[1]
+	ins	v13.s[3], v7.s[2]
+	ins	v15.s[3], v7.s[3]
+
+	st1	{v8.4s, v9.4s, v10.4s, v11.4s}, [B00], #64
+	st1	{v12.4s, v13.4s, v14.4s, v15.4s}, [B00], #64
+.endm
+
+.macro COPY2x8
+	ldr	d0, [A01], #8
+	ldr	d1, [A02], #8
+	ins	v8.s[0], v0.s[0]
+	ins	v10.s[0], v0.s[1]
+	ins	v8.s[1], v1.s[0]
+	ins	v10.s[1], v1.s[1]
+
+	ldr	d2, [A03], #8
+	ldr	d3, [A04], #8
+	ins	v8.s[2], v2.s[0]
+	ins	v10.s[2], v2.s[1]
+	ins	v8.s[3], v3.s[0]
+	ins	v10.s[3], v3.s[1]
+
+	ldr	d4, [A05], #8
+	ldr	d5, [A06], #8
+	ins	v9.s[0], v4.s[0]
+	ins	v11.s[0], v4.s[1]
+	ins	v9.s[1], v5.s[0]
+	ins	v11.s[1], v5.s[1]
+
+	ldr	d6, [A07], #8
+	ldr	d7, [A08], #8
+	ins	v9.s[2], v6.s[0]
+	ins	v11.s[2], v6.s[1]
+	ins	v9.s[3], v7.s[0]
+	ins	v11.s[3], v7.s[1]
+
+	st1	{v8.4s, v9.4s, v10.4s, v11.4s}, [B00], #64
+.endm
+
+.macro COPY1x8
+	ldr	s0, [A01], #4
+	ldr	s1, [A02], #4
+	ins	v8.s[0], v0.s[0]
+	ins	v8.s[1], v1.s[0]
+
+	ldr	s2, [A03], #4
+	ldr	s3, [A04], #4
+	ins	v8.s[2], v2.s[0]
+	ins	v8.s[3], v3.s[0]
+
+	ldr	s4, [A05], #4
+	ldr	s5, [A06], #4
+	ins	v9.s[0], v4.s[0]
+	ins	v9.s[1], v5.s[0]
+
+	ldr	s6, [A07], #4
+	ldr	s7, [A08], #4
+	ins	v9.s[2], v6.s[0]
+	ins	v9.s[3], v7.s[0]
+
+	st1	{v8.4s, v9.4s}, [B00], #32
+.endm
+
+.macro COPY4x4
+	ldr	q0, [A01], #16
+	ldr	q1, [A02], #16
+	ins	v8.s[0], v0.s[0]
+	ins	v9.s[0], v0.s[1]
+	ins	v10.s[0], v0.s[2]
+	ins	v11.s[0], v0.s[3]
+	ins	v8.s[1], v1.s[0]
+	ins	v9.s[1], v1.s[1]
+	ins	v10.s[1], v1.s[2]
+	ins	v11.s[1], v1.s[3]
+
+	ldr	q2, [A03], #16
+	ldr	q3, [A04], #16
+	ins	v8.s[2], v2.s[0]
+	ins	v9.s[2], v2.s[1]
+	ins	v10.s[2], v2.s[2]
+	ins	v11.s[2], v2.s[3]
+	ins	v8.s[3], v3.s[0]
+	ins	v9.s[3], v3.s[1]
+	ins	v10.s[3], v3.s[2]
+	ins	v11.s[3], v3.s[3]
+
+	st1	{v8.4s, v9.4s, v10.4s, v11.4s}, [B00], #64
+.endm
+
+.macro COPY2x4
+	ldr	d0, [A01], #8
+	ldr	d1, [A02], #8
+	ins	v8.s[0], v0.s[0]
+	ins	v9.s[0], v0.s[1]
+	ins	v8.s[1], v1.s[0]
+	ins	v9.s[1], v1.s[1]
+
+	ldr	d2, [A03], #8
+	ldr	d3, [A04], #8
+	ins	v8.s[2], v2.s[0]
+	ins	v9.s[2], v2.s[1]
+	ins	v8.s[3], v3.s[0]
+	ins	v9.s[3], v3.s[1]
+
+	st1	{v8.4s, v9.4s}, [B00], #32
+.endm
+
+.macro COPY1x4
+	ldr	s0, [A01], #4
+	ldr	s1, [A02], #4
+	ins	v8.s[0], v0.s[0]
+	ins	v8.s[1], v1.s[0]
+
+	ldr	s2, [A03], #4
+	ldr	s3, [A04], #4
+	ins	v8.s[2], v2.s[0]
+	ins	v8.s[3], v3.s[0]
+
+	st1	{v8.4s}, [B00], #16
+.endm
+
+.macro COPY4x2
+	ldr	q0, [A01], #16
+	ldr	q1, [A02], #16
+	ins	v8.s[0], v0.s[0]
+	ins	v9.s[0], v0.s[1]
+	ins	v10.s[0], v0.s[2]
+	ins	v11.s[0], v0.s[3]
+	ins	v8.s[1], v1.s[0]
+	ins	v9.s[1], v1.s[1]
+	ins	v10.s[1], v1.s[2]
+	ins	v11.s[1], v1.s[3]
+
+	st1	{v8.2s, v9.2s, v10.2s, v11.2s}, [B00], #32
+.endm
+
+.macro COPY2x2
+	ldr	d0, [A01], #8
+	ldr	d1, [A02], #8
+	ins	v8.s[0], v0.s[0]
+	ins	v9.s[0], v0.s[1]
+	ins	v8.s[1], v1.s[0]
+	ins	v9.s[1], v1.s[1]
+
+	st1	{v8.2s, v9.2s}, [B00], #16
+.endm
+
+.macro COPY1x2
+	ldr	s0, [A01], #4
+	ldr	s1, [A02], #4
+	ins	v8.s[0], v0.s[0]
+	ins	v8.s[1], v1.s[0]
+
+	st1	{v8.2s}, [B00], #8
+.endm
+
+.macro COPY1x1
+	ldr	s0, [A01], #4
+	str	s0, [B00], #4
+.endm
+
+/**************************************************************************************
+* End of macro definitions
+**************************************************************************************/
+
+	PROLOGUE
+
+	.align 5
+
+	SAVE_REGS
+
+	lsl	LDA, LDA, #2					// LDA = LDA * SIZE
+
+.Lsgemm_ncopy_L8_BEGIN:
+
+	asr	J, N, #3					// J = N / 8
+	cmp 	J, #0
+	ble	.Lsgemm_ncopy_L4_BEGIN
+
+	.align	5
+.Lsgemm_ncopy_L8_M4_BEGIN:
+
+	mov	A01, A00
+	add	A02, A01, LDA
+	add	A03, A02, LDA
+	add	A04, A03, LDA
+	add	A05, A04, LDA
+	add	A06, A05, LDA
+	add	A07, A06, LDA
+	add	A08, A07, LDA
+	add	A00, A08, LDA
+
+	asr	I, M, #2					// I = M / 4
+	cmp	I, #0
+	ble	.Lsgemm_ncopy_L8_M4_40
+
+	asr	K, M, #4					// K = M / 16(cacheline)
+	mov	TEMP1, A01
+
+	.align	5
+.Lsgemm_tcopy_L8_warnup_1:
+
+	ldr	s0, [TEMP1], #64
+
+	subs	K, K, #1
+	bgt	.Lsgemm_tcopy_L8_warnup_1
+
+	asr	K, M, #4					// K = M / 16(cacheline)
+	mov	TEMP1, A02
+
+	.align	5
+.Lsgemm_tcopy_L8_warnup_2:
+
+	ldr	s0, [TEMP1], #64
+
+	subs	K, K, #1
+	bgt	.Lsgemm_tcopy_L8_warnup_2
+
+	asr	K, M, #4					// K = M / 16(cacheline)
+	mov	TEMP1, A03
+
+	.align	5
+.Lsgemm_tcopy_L8_warnup_3:
+
+	ldr	s0, [TEMP1], #64
+
+	subs	K, K, #1
+	bgt	.Lsgemm_tcopy_L8_warnup_3
+
+	asr	K, M, #4					// K = M / 16(cacheline)
+	mov	TEMP1, A04
+
+	.align	5
+.Lsgemm_tcopy_L8_warnup_4:
+
+	ldr	s0, [TEMP1], #64
+
+	subs	K, K, #1
+	bgt	.Lsgemm_tcopy_L8_warnup_4
+
+	asr	K, M, #4					// K = M / 16(cacheline)
+	mov	TEMP1, A05
+
+	.align	5
+.Lsgemm_tcopy_L8_warnup_5:
+
+	ldr	s0, [TEMP1], #64
+
+	subs	K, K, #1
+	bgt	.Lsgemm_tcopy_L8_warnup_5
+
+	asr	K, M, #4					// K = M / 16(cacheline)
+	mov	TEMP1, A06
+
+	.align	5
+.Lsgemm_tcopy_L8_warnup_6:
+
+	ldr	s0, [TEMP1], #64
+
+	subs	K, K, #1
+	bgt	.Lsgemm_tcopy_L8_warnup_6
+
+	asr	K, M, #4					// K = M / 16(cacheline)
+	mov	TEMP1, A07
+
+	.align	5
+.Lsgemm_tcopy_L8_warnup_7:
+
+	ldr	s0, [TEMP1], #64
+
+	subs	K, K, #1
+	bgt	.Lsgemm_tcopy_L8_warnup_7
+
+	asr	K, M, #4					// K = M / 16(cacheline)
+	mov	TEMP1, A08
+
+	.align	5
+.Lsgemm_tcopy_L8_warnup_8:
+
+	ldr	s0, [TEMP1], #64
+
+	subs	K, K, #1
+	bgt	.Lsgemm_tcopy_L8_warnup_8
+
+	.align	5
+.Lsgemm_ncopy_L8_M4_20:
+
+	COPY4x8
+
+	subs	I, I, #1
+	bne	.Lsgemm_ncopy_L8_M4_20
+
+.Lsgemm_ncopy_L8_M4_40:
+
+	and	I, M, #2
+	cmp	I, #0
+	ble	.Lsgemm_ncopy_L8_M4_60
+
+	COPY2x8
+
+.Lsgemm_ncopy_L8_M4_60:
+
+	and	I, M, #1
+	cmp	I, #0
+	ble	.Lsgemm_ncopy_L8_M4_END
+
+	COPY1x8
+
+.Lsgemm_ncopy_L8_M4_END:
+
+	subs	J , J, #1						// j--
+	bne	.Lsgemm_ncopy_L8_M4_BEGIN
+
+/*********************************************************************************************/
+
+.Lsgemm_ncopy_L4_BEGIN:
+
+	tst	N, #7
+	ble	.Lsgemm_ncopy_L999
+
+	tst	N, #4
+	ble	.Lsgemm_ncopy_L2_BEGIN
+
+.Lsgemm_ncopy_L4_M4_BEGIN:
+	mov	A01, A00
+	add	A02, A01, LDA
+	add	A03, A02, LDA
+	add	A04, A03, LDA
+	add	A00, A04, LDA
+
+	asr	I, M, #2					// I = M / 4
+	cmp 	I, #0
+	ble	.Lsgemm_ncopy_L4_M4_40
+
+	.align	5
+.Lsgemm_ncopy_L4_M4_20:
+
+	COPY4x4
+
+	subs	I, I, #1
+	bne	.Lsgemm_ncopy_L4_M4_20
+
+.Lsgemm_ncopy_L4_M4_40:
+
+	and	I, M, #2
+	cmp	I, #0
+	ble	.Lsgemm_ncopy_L4_M4_60
+
+	COPY2x4
+
+.Lsgemm_ncopy_L4_M4_60:
+
+	and	I, M, #1
+	cmp	I, #0
+	ble	.Lsgemm_ncopy_L4_M4_END
+
+	COPY1x4
+
+.Lsgemm_ncopy_L4_M4_END:
+
+
+/*********************************************************************************************/
+
+.Lsgemm_ncopy_L2_BEGIN:
+
+	tst	N, #2
+	ble	.Lsgemm_ncopy_L1_BEGIN
+
+.Lsgemm_ncopy_L2_M4_BEGIN:
+
+	mov	A01, A00
+	add	A02, A01, LDA
+	add	A00, A02, LDA
+
+	asr	I, M, #2					// I = M / 4
+	cmp	I, #0
+	ble	.Lsgemm_ncopy_L2_M4_40
+
+	.align	5
+.Lsgemm_ncopy_L2_M4_20:
+
+	COPY4x2
+
+	subs	I , I , #1
+	bne	.Lsgemm_ncopy_L2_M4_20
+
+
+.Lsgemm_ncopy_L2_M4_40:
+
+	and	I, M, #2
+	cmp	I, #0
+	ble	.Lsgemm_ncopy_L2_M4_60
+
+	COPY2x2
+
+.Lsgemm_ncopy_L2_M4_60:
+
+	and	I, M, #1
+	cmp	I, #0
+	ble	.Lsgemm_ncopy_L2_M4_END
+
+	COPY1x2
+
+.Lsgemm_ncopy_L2_M4_END:
+
+.Lsgemm_ncopy_L1_BEGIN:
+
+	tst	N, #1
+	ble	.Lsgemm_ncopy_L999
+
+.Lsgemm_ncopy_L1_M1_BEGIN:
+
+	mov	A01, A00
+
+	mov	I, M
+	cmp	I, #0
+	ble	.Lsgemm_ncopy_L1_M1_END
+
+	.align	5
+.Lsgemm_ncopy_L1_M1_20:
+
+	COPY1x1
+
+	subs	I, I, #1
+	bne	.Lsgemm_ncopy_L1_M1_20
+
+.Lsgemm_ncopy_L1_M1_END:
+
+.Lsgemm_ncopy_L999:
+
+	mov	x0, #0
+	RESTORE_REGS
+	ret
+
+	EPILOGUE
diff --git a/kernel/arm64/sgemm_tcopy_8.S b/kernel/arm64/sgemm_tcopy_8.S
new file mode 100644
index 000000000..7d81ba266
--- /dev/null
+++ b/kernel/arm64/sgemm_tcopy_8.S
@@ -0,0 +1,707 @@
+/***************************************************************************
+Copyright (c) 2016, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A00 PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+#define ASSEMBLER
+#include "common.h"
+
+#define	M		x0
+#define	N		x1
+#define	A		x2
+#define	LDA		x3
+#define	B		x4
+
+#define M8		x5
+
+#define	A01		x6
+#define	A02		x7
+#define	A03		x8
+#define	A04		x9
+#define	A05		x10
+#define	A06		x11
+#define	A07		x12
+#define	A08		x13
+
+#define	B01		x14
+#define	B02		x15
+#define	B03		x16
+#define	B04		x17
+#define	B00		x22
+
+
+#define I		x18
+#define	J		x19
+
+#define TEMP1		x20
+
+#define A_PREFETCH	256
+
+/**************************************************************************************
+* Macro definitions
+**************************************************************************************/
+.macro SAVE_REGS
+	add	sp, sp, #-(11 * 16)
+	stp	d8, d9, [sp, #(0 * 16)]
+	stp	d10, d11, [sp, #(1 * 16)]
+	stp	d12, d13, [sp, #(2 * 16)]
+	stp	d14, d15, [sp, #(3 * 16)]
+	stp	d16, d17, [sp, #(4 * 16)]
+	stp	x18, x19, [sp, #(5 * 16)]
+	stp	x20, x21, [sp, #(6 * 16)]
+	stp	x22, x23, [sp, #(7 * 16)]
+	stp	x24, x25, [sp, #(8 * 16)]
+	stp	x26, x27, [sp, #(9 * 16)]
+	str	x28, [sp, #(10 * 16)]
+.endm
+
+.macro RESTORE_REGS
+	ldp	d8, d9, [sp, #(0 * 16)]
+	ldp	d10, d11, [sp, #(1 * 16)]
+	ldp	d12, d13, [sp, #(2 * 16)]
+	ldp	d14, d15, [sp, #(3 * 16)]
+	ldp	d16, d17, [sp, #(4 * 16)]
+	ldp	x18, x19, [sp, #(5 * 16)]
+	ldp	x20, x21, [sp, #(6 * 16)]
+	ldp	x22, x23, [sp, #(7 * 16)]
+	ldp	x24, x25, [sp, #(8 * 16)]
+	ldp	x26, x27, [sp, #(9 * 16)]
+	ldr	x28, [sp, #(10 * 16)]
+	add	sp, sp, #(11*16)
+.endm
+
+/*************************************************************************************************************************/
+
+.macro COPY8x8
+	prfm	PLDL1KEEP, [A01, #A_PREFETCH]
+	prfm	PLDL1KEEP, [A02, #A_PREFETCH]
+	prfm	PLDL1KEEP, [A03, #A_PREFETCH]
+	prfm	PLDL1KEEP, [A04, #A_PREFETCH]
+	prfm	PLDL1KEEP, [A05, #A_PREFETCH]
+	prfm	PLDL1KEEP, [A06, #A_PREFETCH]
+	prfm	PLDL1KEEP, [A07, #A_PREFETCH]
+	prfm	PLDL1KEEP, [A08, #A_PREFETCH]
+
+	ldp	q0, q1, [A01]
+	ldp	q2, q3, [A02]
+	add	A01, A01, #32
+	add	A02, A02, #32
+
+	st1	{v0.4s, v1.4s, v2.4s, v3.4s}, [B00]
+	add	TEMP1, B00, #64
+
+	ldp	q4, q5, [A03]
+	ldp	q6, q7, [A04]
+	add	A03, A03, #32
+	add	A04, A04, #32
+
+	st1	{v4.4s, v5.4s, v6.4s, v7.4s}, [TEMP1]
+	add	TEMP1, TEMP1, #64
+
+	ldp	q8, q9, [A05]
+	ldp	q10, q11, [A06]
+	add	A05, A05, #32
+	add	A06, A06, #32
+
+	st1	{v8.4s, v9.4s, v10.4s, v11.4s}, [TEMP1]
+	add	TEMP1, TEMP1, #64
+
+	ldp	q12, q13, [A07]
+	ldp	q14, q15, [A08]
+	add	A07, A07, #32
+	add	A08, A08, #32
+
+	st1	{v12.4s, v13.4s, v14.4s, v15.4s}, [TEMP1]
+	add	TEMP1, TEMP1, #64
+
+	add	B00, B00, M8
+.endm
+
+.macro COPY4x8
+	//prfm	PLDL1KEEP, [A01, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A02, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A03, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A04, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A05, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A06, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A07, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A08, #A_PREFETCH]
+
+	ldr	q0, [A01]
+	ldr	q1, [A02]
+	ldr	q2, [A03]
+	ldr	q3, [A04]
+	add	A01, A01, #16
+	add	A02, A02, #16
+	add	A03, A03, #16
+	add	A04, A04, #16
+
+	st1	{v0.4s, v1.4s, v2.4s, v3.4s}, [B01]
+	add	B01, B01, #64
+
+	ldr	q4, [A05]
+	ldr	q5, [A06]
+	ldr	q6, [A07]
+	ldr	q7, [A08]
+
+	add	A05, A05, #16
+	add	A06, A06, #16
+	add	A07, A07, #16
+	add	A08, A08, #16
+
+	st1	{v4.4s, v5.4s, v6.4s, v7.4s}, [B01]
+	add	B01, B01, #64
+.endm
+
+.macro COPY2x8
+	//prfm	PLDL1KEEP, [A01, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A02, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A03, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A04, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A05, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A06, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A07, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A08, #A_PREFETCH]
+
+	ldr	d0, [A01]
+	ldr	d1, [A02]
+	ldr	d2, [A03]
+	ldr	d3, [A04]
+
+	add	A01, A01, #8
+	add	A02, A02, #8
+	add	A03, A03, #8
+	add	A04, A04, #8
+
+	stp	d0, d1, [B02]
+	add	B02, B02, #16
+	stp	d2, d3, [B02]
+	add	B02, B02, #16
+
+	ldr	d4, [A05]
+	ldr	d5, [A06]
+	ldr	d6, [A07]
+	ldr	d7, [A08]
+
+	add	A05, A05, #8
+	add	A06, A06, #8
+	add	A07, A07, #8
+	add	A08, A08, #8
+
+	stp	d4, d5, [B02]
+	add	B02, B02, #16
+	stp	d6, d7, [B02]
+	add	B02, B02, #16
+
+.endm
+
+.macro COPY1x8
+	//prfm	PLDL1KEEP, [A01, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A02, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A03, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A04, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A05, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A06, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A07, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A08, #A_PREFETCH]
+
+	ldr	s0, [A01]
+	ldr	s1, [A02]
+	ldr	s2, [A03]
+	ldr	s3, [A04]
+
+	add	A01, A01, #4
+	add	A02, A02, #4
+	add	A03, A03, #4
+	add	A04, A04, #4
+
+	stp	s0, s1, [B03]
+	add	B03, B03, #8
+	stp	s2, s3, [B03]
+	add	B03, B03, #8
+
+	ldr	s4, [A05]
+	ldr	s5, [A06]
+	ldr	s6, [A07]
+	ldr	s7, [A08]
+
+	ldr	d4, [A05], #8
+	ldr	d5, [A06], #8
+	ldr	d6, [A07], #8
+	ldr	d7, [A08], #8
+
+	stp	s4, s5, [B03]
+	add	B03, B03, #8
+	stp	s6, s7, [B03]
+	add	B03, B03, #8
+
+.endm
+
+/*************************************************************************************************************************/
+
+.macro COPY8x4
+	prfm	PLDL1KEEP, [A01, #A_PREFETCH]
+	prfm	PLDL1KEEP, [A02, #A_PREFETCH]
+	prfm	PLDL1KEEP, [A03, #A_PREFETCH]
+	prfm	PLDL1KEEP, [A04, #A_PREFETCH]
+
+	ldp	q0, q1, [A01]
+	ldp	q2, q3, [A02]
+	add	A01, A01, #32
+	add	A02, A02, #32
+
+	st1	{v0.4s, v1.4s, v2.4s, v3.4s}, [B00]
+	add	TEMP1, B00, #64
+
+	ldp	q4, q5, [A03]
+	ldp	q6, q7, [A04]
+	add	A03, A03, #32
+	add	A04, A04, #32
+
+	st1	{v4.4s, v5.4s, v6.4s, v7.4s}, [TEMP1]
+	add	TEMP1, TEMP1, #64
+
+	add	B00, B00, M8
+.endm
+
+.macro COPY4x4
+	//prfm	PLDL1KEEP, [A01, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A02, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A03, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A04, #A_PREFETCH]
+
+	ldr	q0, [A01]
+	ldr	q1, [A02]
+	ldr	q2, [A03]
+	ldr	q3, [A04]
+	add	A01, A01, #16
+	add	A02, A02, #16
+	add	A03, A03, #16
+	add	A04, A04, #16
+
+	st1	{v0.4s, v1.4s, v2.4s, v3.4s}, [B01]
+
+	add	B01, B01, #64
+.endm
+
+.macro COPY2x4
+	//prfm	PLDL1KEEP, [A01, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A02, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A03, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A04, #A_PREFETCH]
+
+	ldr	d0, [A01]
+	ldr	d1, [A02]
+	ldr	d2, [A03]
+	ldr	d3, [A04]
+
+	add	A01, A01, #8
+	add	A02, A02, #8
+	add	A03, A03, #8
+	add	A04, A04, #8
+
+	stp	d0, d1, [B02]
+	add	B02, B02, #16
+	stp	d2, d3, [B02]
+
+	add	B02, B02, #16
+.endm
+
+.macro COPY1x4
+	//prfm	PLDL1KEEP, [A01, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A02, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A03, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A04, #A_PREFETCH]
+
+	ldr	s0, [A01]
+	ldr	s1, [A02]
+	ldr	s2, [A03]
+	ldr	s3, [A04]
+
+	add	A01, A01, #4
+	add	A02, A02, #4
+	add	A03, A03, #4
+	add	A04, A04, #4
+
+	stp	s0, s1, [B03]
+	add	B03, B03, #8
+	stp	s2, s3, [B03]
+	add	B03, B03, #8
+
+.endm
+
+/*************************************************************************************************************************/
+
+.macro COPY8x2
+	prfm	PLDL1KEEP, [A01, #A_PREFETCH]
+	prfm	PLDL1KEEP, [A02, #A_PREFETCH]
+
+	ld1	{v0.4s, v1.4s}, [A01]
+	ld1	{v2.4s, v3.4s}, [A02]
+	add	A01, A01, #32
+	add	A02, A02, #32
+
+	st1	{v0.4s, v1.4s, v2.4s, v3.4s}, [B00]
+	add	B00, B00, M8
+.endm
+
+.macro COPY4x2
+	//prfm	PLDL1KEEP, [A01, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A02, #A_PREFETCH]
+
+	ldr	q0, [A01]
+	ldr	q1, [A02]
+	add	A01, A01, #16
+	add	A02, A02, #16
+
+	stp	q0, q1, [B01]
+	add	B01, B01, #32
+.endm
+
+.macro COPY2x2
+	//prfm	PLDL1KEEP, [A01, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A02, #A_PREFETCH]
+
+	ldr	d0, [A01]
+	ldr	d1, [A02]
+
+	add	A01, A01, #8
+	add	A02, A02, #8
+
+	stp	d0, d1, [B02]
+	add	B02, B02, #16
+.endm
+
+.macro COPY1x2
+	//prfm	PLDL1KEEP, [A01, #A_PREFETCH]
+	//prfm	PLDL1KEEP, [A02, #A_PREFETCH]
+
+	ldr	s0, [A01]
+	ldr	s1, [A02]
+
+	add	A01, A01, #4
+	add	A02, A02, #4
+
+	stp	s0, s1, [B03]
+
+	add	B03, B03, #8
+.endm
+
+/*************************************************************************************************************************/
+
+.macro COPY8x1
+	prfm	PLDL1KEEP, [A01, #A_PREFETCH]
+
+	ldp	q0, q1, [A01]
+	add	A01, A01, #32
+	stp	q0, q1, [B00]
+
+	add	B00, B00, M8
+.endm
+
+.macro COPY4x1
+	//prfm	PLDL1KEEP, [A01, #A_PREFETCH]
+
+	ldr q0, [A01]
+	add	A01, A01, #16
+	str q0, [B01]
+
+	add	B01, B01, #16
+.endm
+
+.macro COPY2x1
+	//prfm	PLDL1KEEP, [A01, #A_PREFETCH]
+
+	ldr	d0, [A01]
+	add	A01, A01, #8
+	str d0, [B02]
+
+	add	B02, B02, #8
+.endm
+
+.macro COPY1x1
+	//prfm	PLDL1KEEP, [A01, #A_PREFETCH]
+
+	ldr	s0, [A01]
+	add	A01, A01, #4
+	str	s0, [B03]
+
+	add	B03, B03, #4
+.endm
+
+/**************************************************************************************
+* End of macro definitions
+**************************************************************************************/
+
+	PROLOGUE
+
+	.align 5
+
+	SAVE_REGS
+
+	lsl	LDA, LDA, #2					// LDA = LDA * SIZE
+
+	lsl	TEMP1, M, #2					// TEMP1 = M * SIZE
+
+	and	B01 , N , #-8
+	and	B02 , N , #-4
+	and	B03 , N , #-2
+
+	mul	B01, B01, TEMP1
+	mul	B02, B02, TEMP1
+	mul	B03, B03, TEMP1
+
+	add	B01 , B01, B
+	add	B02 , B02, B
+	add	B03 , B03, B
+
+	lsl	M8, M, #5					// M8 = M * 8 * SIZE
+
+.Lsgemm_tcopy_L8_BEGIN:
+
+	asr 	J, M, #3					// J = M / 8
+	cmp 	J, #0
+	ble	.Lsgemm_tcopy_L4_BEGIN
+
+	.align	5
+.Lsgemm_tcopy_L8_M8_BEGIN:
+
+	mov	A01, A
+	add	A02, A01, LDA
+	add	A03, A02, LDA
+	add	A04, A03, LDA
+	add	A05, A04, LDA
+	add	A06, A05, LDA
+	add	A07, A06, LDA
+	add	A08, A07, LDA
+	add	A, A08, LDA
+
+	mov	B00, B
+	add	B, B00, #256					// B = B + 8 * 8 * SIZE
+
+	asr	I, N, #3					// I = N / 8
+	cmp 	I, #0
+	ble	.Lsgemm_tcopy_L8_M8_40
+
+	.align	5
+.Lsgemm_tcopy_L8_M8_20:
+
+	COPY8x8
+
+	subs	I , I , #1
+	bne	.Lsgemm_tcopy_L8_M8_20
+
+.Lsgemm_tcopy_L8_M8_40:
+
+	tst	N , #4
+	ble	.Lsgemm_tcopy_L8_M8_60
+
+	COPY4x8
+
+.Lsgemm_tcopy_L8_M8_60:
+
+	tst	N , #2
+	ble	.Lsgemm_tcopy_L8_M8_80
+
+	COPY2x8
+
+.Lsgemm_tcopy_L8_M8_80:
+
+	tst	N, #1
+	ble	.Lsgemm_tcopy_L8_M8_END
+
+	COPY1x8
+
+.Lsgemm_tcopy_L8_M8_END:
+
+	subs	J, J, #1						// j--
+	bne	.Lsgemm_tcopy_L8_M8_BEGIN
+
+/*********************************************************************************************/
+
+.Lsgemm_tcopy_L4_BEGIN:
+
+	tst	M, #7
+	ble	.Lsgemm_tcopy_L999
+
+	tst	M, #4
+	ble	.Lsgemm_tcopy_L2_BEGIN
+
+.Lsgemm_tcopy_L4_M8_BEGIN:
+
+	mov	A01, A
+	add	A02, A01, LDA
+	add	A03, A02, LDA
+	add	A04, A03, LDA
+	add	A, A04, LDA
+
+	mov	B00, B
+	add	B, B00, #128					// B = B + 4 * 8 * SIZE
+
+	asr	I, N, #3					// I = N / 8
+	cmp 	I, #0
+	ble	.Lsgemm_tcopy_L4_M8_40
+
+	.align	5
+.Lsgemm_tcopy_L4_M8_20:
+
+	COPY8x4
+
+	subs	I , I , #1
+	bne	.Lsgemm_tcopy_L4_M8_20
+
+.Lsgemm_tcopy_L4_M8_40:
+
+	tst	N , #4
+	ble	.Lsgemm_tcopy_L4_M8_60
+
+	COPY4x4
+
+.Lsgemm_tcopy_L4_M8_60:
+
+	tst	N , #2
+	ble	.Lsgemm_tcopy_L4_M8_80
+
+	COPY2x4
+
+.Lsgemm_tcopy_L4_M8_80:
+
+	tst	N , #1
+	ble	.Lsgemm_tcopy_L4_M8_END
+
+	COPY1x4
+
+
+.Lsgemm_tcopy_L4_M8_END:
+
+/*********************************************************************************************/
+
+.Lsgemm_tcopy_L2_BEGIN:
+
+	tst	M, #3
+	ble	.Lsgemm_tcopy_L999
+
+	tst	M, #2
+	ble	.Lsgemm_tcopy_L1_BEGIN
+
+.Lsgemm_tcopy_L2_M16_BEGIN:
+
+	mov	A01, A
+	add	A02, A01, LDA
+	add	A, A02, LDA
+
+	mov	B00, B
+	add	B, B00, #64					// B = B + 2 * 8 * SIZE
+
+	asr	I, N, #3					// I = N / 8
+	cmp 	I, #0
+	ble	.Lsgemm_tcopy_L2_M8_40
+
+	.align	5
+.Lsgemm_tcopy_L2_M8_20:
+
+	COPY8x2
+
+	subs	I , I , #1
+	bne	.Lsgemm_tcopy_L2_M8_20
+
+.Lsgemm_tcopy_L2_M8_40:
+
+	tst	N , #4
+	ble	.Lsgemm_tcopy_L2_M8_60
+
+	COPY4x2
+
+.Lsgemm_tcopy_L2_M8_60:
+
+	tst	N , #2
+	ble	.Lsgemm_tcopy_L2_M8_80
+
+	COPY2x2
+
+.Lsgemm_tcopy_L2_M8_80:
+
+	tst	N , #1
+	ble	.Lsgemm_tcopy_L2_M8_END
+
+	COPY1x2
+
+.Lsgemm_tcopy_L2_M8_END:
+
+/*********************************************************************************************/
+
+.Lsgemm_tcopy_L1_BEGIN:
+
+	tst	M, #1
+	ble	.Lsgemm_tcopy_L999
+
+
+.Lsgemm_tcopy_L1_M16_BEGIN:
+
+	mov	A01, A						// A01 = A
+	mov	B00, B
+
+	asr	I, N, #3					// I = M / 8
+	cmp 	I, #0
+	ble	.Lsgemm_tcopy_L1_M8_40
+
+	.align	5
+.Lsgemm_tcopy_L1_M8_20:
+
+	COPY8x1
+
+	subs	I , I , #1
+	bne	.Lsgemm_tcopy_L1_M8_20
+
+.Lsgemm_tcopy_L1_M8_40:
+
+	tst	N , #4
+	ble	.Lsgemm_tcopy_L1_M8_60
+
+	COPY4x1
+
+.Lsgemm_tcopy_L1_M8_60:
+
+	tst	N , #2
+	ble	.Lsgemm_tcopy_L1_M8_80
+
+	COPY2x1
+
+.Lsgemm_tcopy_L1_M8_80:
+
+	tst	N , #1
+	ble	.Lsgemm_tcopy_L1_M8_END
+
+	COPY1x1
+
+
+.Lsgemm_tcopy_L1_M8_END:
+
+.Lsgemm_tcopy_L999:
+
+	mov	x0, #0						// set return value
+	RESTORE_REGS
+	ret
+
+	EPILOGUE
diff --git a/kernel/arm64/strmm_kernel_8x8_cortexa53.S b/kernel/arm64/strmm_kernel_8x8_cortexa53.S
new file mode 100644
index 000000000..4b84623f3
--- /dev/null
+++ b/kernel/arm64/strmm_kernel_8x8_cortexa53.S
@@ -0,0 +1,2823 @@
+/*******************************************************************************
+Copyright (c) 2015, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*******************************************************************************/
+#define ASSEMBLER
+#include "common.h"
+
+/*                   X0          X1          X2          s0        X3        x4       x5           x6               x7 */
+/*int CNAME(BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alpha,FLOAT* ba,FLOAT* bb,FLOAT* C,BLASLONG ldc, BLASLONG offset) */
+
+#define origM		x0
+#define origN		x1
+#define origK		x2
+#define origPA		x3
+#define origPB		x4
+#define pC		x5
+#define LDC		x6
+#define offset		x7
+#define counterL	x8
+#define counterI	x9
+#define counterJ	x10
+#define pB		x11
+#define pCRow0		x12
+#define pCRow1		x13
+#define pCRow2		x14
+#define pA		x15
+#define temp		x16
+#define tempOffset	x17
+#define tempK		x18
+
+#define alpha0		s10
+#define alphaV0		v10.s[0]
+#define alpha1		s11
+#define alphaV1		v11.s[0]
+#define alpha2		s14
+#define alphaV2		v14.s[0]
+#define alpha3		s15
+#define alphaV3		v15.s[0]
+
+// 00 origM
+// 01 origN
+// 02 origK
+// 03 origPA
+// 04 origPB
+// 05 pC
+// 06 origLDC -> LDC
+// 07 offset
+// 08 counterL
+// 09 counterI
+// 10 counterJ
+// 11 pB
+// 12 pCRow0
+// 13 pCRow1
+// 14 pCRow2
+// 15 pA
+// 16 temp
+// 17 tempOffset
+// 18 must save tempK
+// 19 must save
+// 20 must save pA0_2, pA0_3
+// 21 must save pA0_6, pA0_7
+// 22 must save pA1_2, pA1_3
+// 23 must save pA1_6, pA1_7
+// 24 must save pB0_2, pB0_3
+// 25 must save pB0_6, pB0_7
+// 26 must save pB1_2, pB1_3
+// 27 must save pB1_6, pB1_7
+// 28 must save
+// 29 frame
+// 30 link
+// 31 sp
+
+//v00 ALPHA -> pA0_0, pA0_1, pA0_2, pA0_3
+//v01 pA0_4, pA0_5, pA0_6, pA0_7
+//v02 pA1_0, pA1_1, pA1_2, pA1_3
+//v03 pA1_4, pA1_5, pA1_6, pA1_7
+//v04 pB0_0, pB0_1, pB0_2, pB0_3
+//v05 pB0_4, pB0_5, pB0_6, pB0_7
+//v06 pB1_0, pB1_1, pB1_2, pB1_3
+//v07 pB1_4, pB1_5, pB1_6, pB1_7
+//v08 must save
+//v09 must save
+//v10 must save ALPHA0
+//v11 must save ALPHA1
+//v12 must save
+//v13 must save
+//v14 must save ALPHA2
+//v15 must save ALPHA3
+//v16 must save C00, C01, C02, C03
+//v17 must save C04, C05, C06, C07
+//v18 C08, C09, C10, C11
+//v19 C12, C13, C14, C15
+//v20 C16, C17, C18, C19
+//v21 C20, C21, C22, C23
+//v22 C24, C25, C26, C27
+//v23 C28, C29, C30, C31
+//v24 C32, C33, C34, C35
+//v25 C36, C37, C38, C39
+//v26 C40, C41, C42, C43
+//v27 C44, C45, C46, C47
+//v28 C48, C49, C50, C51
+//v29 C52, C53, C54, C55
+//v30 C56, C57, C58, C59
+//v31 C60, C61, C62, C63
+
+/*******************************************************************************
+* Macro definitions
+*******************************************************************************/
+
+.macro INIT8x8
+	fmov		s16, wzr
+	fmov		s17, wzr
+	fmov		s18, s16
+	fmov		s19, s17
+	fmov		s20, wzr
+	fmov		s21, s16
+	fmov		s22, s17
+	fmov		s23, s18
+	fmov		s24, wzr
+	fmov		s25, s16
+	fmov		s26, s17
+	fmov		s27, s18
+	fmov		s28, wzr
+	fmov		s29, s16
+	fmov		s30, s17
+	fmov		s31, s18
+.endm
+
+.macro KERNEL8x8_I
+	ld1	{v0.4s, v1.4s}, [pA], #32
+	ld1	{v4.4s, v5.4s}, [pB], #32
+	ldr	d2, [pA], #8
+	ldr	d6, [pB], #8
+	ldr	d3, [pA, #8]
+	ldr	d7, [pB, #8]
+
+	ldr	x22, [pA], #16
+	fmul	v16.4s, v0.4s, v4.s[0]
+	ldr	x26, [pB], #16
+	fmul	v17.4s, v1.4s, v4.s[0]
+	ldr	x23, [pA], #8
+	fmul	v18.4s, v0.4s, v4.s[1]
+	ldr	x27, [pB], #8
+	fmul	v19.4s, v1.4s, v4.s[1]
+	fmul	v20.4s, v0.4s, v4.s[2]
+	fmul	v21.4s, v1.4s, v4.s[2]
+	fmul	v22.4s, v0.4s, v4.s[3]
+	fmul	v23.4s, v1.4s, v4.s[3]
+	fmul	v24.4s, v0.4s, v5.s[0]
+	fmul	v25.4s, v1.4s, v5.s[0]
+	fmul	v26.4s, v0.4s, v5.s[1]
+	fmul	v27.4s, v1.4s, v5.s[1]
+	fmul	v28.4s, v0.4s, v5.s[2]
+	fmul	v29.4s, v1.4s, v5.s[2]
+	fmul	v30.4s, v0.4s, v5.s[3]
+	fmul	v31.4s, v1.4s, v5.s[3]
+.endm
+
+.macro KERNEL8x8_M1
+	ldr	d2, [pA], #8
+	fmov	v0.d[1], x20
+	ldr	d6, [pB], #8
+	fmov	v4.d[1], x24
+	ldr	d3, [pA, #8]
+	fmov	v1.d[1], x21
+	ldr	d7, [pB, #8]
+	fmov	v5.d[1], x25
+	fmla	v16.4s, v0.4s, v4.s[0]
+	ldr	x22, [pA], #16
+	fmla	v17.4s, v1.4s, v4.s[0]
+	ldr	x26, [pB], #16
+	fmla	v18.4s, v0.4s, v4.s[1]
+	ldr	x23, [pA], #8
+	fmla	v19.4s, v1.4s, v4.s[1]
+	ldr	x27, [pB], #8
+	fmla	v20.4s, v0.4s, v4.s[2]
+	fmla	v21.4s, v1.4s, v4.s[2]
+	fmla	v22.4s, v0.4s, v4.s[3]
+	fmla	v23.4s, v1.4s, v4.s[3]
+	fmla	v24.4s, v0.4s, v5.s[0]
+	fmla	v25.4s, v1.4s, v5.s[0]
+	fmla	v26.4s, v0.4s, v5.s[1]
+	fmla	v27.4s, v1.4s, v5.s[1]
+	fmla	v28.4s, v0.4s, v5.s[2]
+	fmla	v29.4s, v1.4s, v5.s[2]
+	fmla	v30.4s, v0.4s, v5.s[3]
+	fmla	v31.4s, v1.4s, v5.s[3]
+.endm
+
+.macro KERNEL8x8_M2
+	ldr	d0, [pA], #8
+	fmov	v2.d[1], x22
+	ldr	d4, [pB], #8
+	fmov	v6.d[1], x26
+	ldr	d1, [pA, #8]
+	fmov	v3.d[1], x23
+	ldr	d5, [pB, #8]
+	fmov	v7.d[1], x27
+	fmla	v16.4s, v2.4s, v6.s[0]
+	ldr	x20, [pA], #16
+	fmla	v17.4s, v3.4s, v6.s[0]
+	ldr	x24, [pB], #16
+	fmla	v18.4s, v2.4s, v6.s[1]
+	ldr	x21, [pA], #8
+	fmla	v19.4s, v3.4s, v6.s[1]
+	ldr	x25, [pB], #8
+	fmla	v20.4s, v2.4s, v6.s[2]
+	fmla	v21.4s, v3.4s, v6.s[2]
+	fmla	v22.4s, v2.4s, v6.s[3]
+	fmla	v23.4s, v3.4s, v6.s[3]
+	fmla	v24.4s, v2.4s, v7.s[0]
+	fmla	v25.4s, v3.4s, v7.s[0]
+	fmla	v26.4s, v2.4s, v7.s[1]
+	fmla	v27.4s, v3.4s, v7.s[1]
+	fmla	v28.4s, v2.4s, v7.s[2]
+	fmla	v29.4s, v3.4s, v7.s[2]
+	fmla	v30.4s, v2.4s, v7.s[3]
+	fmla	v31.4s, v3.4s, v7.s[3]
+.endm
+
+.macro KERNEL8x8_E
+	fmov	v2.d[1], x22
+	fmov	v6.d[1], x26
+	fmov	v3.d[1], x23
+	fmov	v7.d[1], x27
+	fmla	v16.4s, v2.4s, v6.s[0]
+	fmla	v17.4s, v3.4s, v6.s[0]
+	fmla	v18.4s, v2.4s, v6.s[1]
+	fmla	v19.4s, v3.4s, v6.s[1]
+	fmla	v20.4s, v2.4s, v6.s[2]
+	fmla	v21.4s, v3.4s, v6.s[2]
+	fmla	v22.4s, v2.4s, v6.s[3]
+	fmla	v23.4s, v3.4s, v6.s[3]
+	fmla	v24.4s, v2.4s, v7.s[0]
+	fmla	v25.4s, v3.4s, v7.s[0]
+	fmla	v26.4s, v2.4s, v7.s[1]
+	fmla	v27.4s, v3.4s, v7.s[1]
+	fmla	v28.4s, v2.4s, v7.s[2]
+	fmla	v29.4s, v3.4s, v7.s[2]
+	fmla	v30.4s, v2.4s, v7.s[3]
+	fmla	v31.4s, v3.4s, v7.s[3]
+.endm
+
+.macro KERNEL8x8_SUB
+	ld1	{v4.4s}, [pB]
+	add	pB, pB, #16
+	ld1	{v5.4s}, [pB]
+	add	pB, pB, #16
+	ld1	{v0.4s}, [pA]
+	add	pA, pA, #16
+	ld1	{v1.4s}, [pA]
+	add	pA, pA, #16
+
+	fmla	v16.4s, v0.4s, v4.s[0]
+	fmla	v17.4s, v1.4s, v4.s[0]
+	fmla	v18.4s, v0.4s, v4.s[1]
+	fmla	v19.4s, v1.4s, v4.s[1]
+	fmla	v20.4s, v0.4s, v4.s[2]
+	fmla	v21.4s, v1.4s, v4.s[2]
+	fmla	v22.4s, v0.4s, v4.s[3]
+	fmla	v23.4s, v1.4s, v4.s[3]
+	fmla	v24.4s, v0.4s, v5.s[0]
+	fmla	v25.4s, v1.4s, v5.s[0]
+	fmla	v26.4s, v0.4s, v5.s[1]
+	fmla	v27.4s, v1.4s, v5.s[1]
+	fmla	v28.4s, v0.4s, v5.s[2]
+	fmla	v29.4s, v1.4s, v5.s[2]
+	fmla	v30.4s, v0.4s, v5.s[3]
+	fmla	v31.4s, v1.4s, v5.s[3]
+.endm
+
+.macro SAVE8x8
+	add	pCRow1, pCRow0, LDC
+
+	fmul	v0.4s, v16.4s, alphaV0
+	fmul	v1.4s, v17.4s, alphaV1
+	st1 	{v0.4s, v1.4s}, [pCRow0]
+
+	add	pCRow2, pCRow1, LDC
+
+	fmul	v2.4s, v18.4s, alphaV2
+	fmul	v3.4s, v19.4s, alphaV3
+	st1 	{v2.4s, v3.4s}, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+	fmul	v4.4s, v20.4s, alphaV0
+	fmul	v5.4s, v21.4s, alphaV1
+	st1 	{v4.4s, v5.4s}, [pCRow2]
+
+	add	pCRow2, pCRow1, LDC
+
+	fmul	v6.4s, v22.4s, alphaV2
+	fmul	v7.4s, v23.4s, alphaV3
+	st1 	{v6.4s, v7.4s}, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+	fmul	v0.4s, v24.4s, alphaV0
+	fmul	v1.4s, v25.4s, alphaV1
+	st1 	{v0.4s, v1.4s}, [pCRow2]
+
+	add	pCRow2, pCRow1, LDC
+
+	fmul	v2.4s, v26.4s, alphaV2
+	fmul	v3.4s, v27.4s, alphaV3
+	st1 	{v2.4s, v3.4s}, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+	fmul	v4.4s, v28.4s, alphaV0
+	fmul	v5.4s, v29.4s, alphaV1
+	st1 	{v4.4s, v5.4s}, [pCRow2]
+
+	fmul	v6.4s, v30.4s, alphaV2
+	fmul	v7.4s, v31.4s, alphaV3
+	st1 	{v6.4s, v7.4s}, [pCRow1]
+
+	add	pCRow0, pCRow0, #32
+.endm
+
+/******************************************************************************/
+
+
+.macro INIT4x8
+	fmov		s16, wzr
+	fmov		s18, wzr
+	fmov		s20, wzr
+	fmov		s22, s16
+	fmov		s24, wzr
+	fmov		s26, s16
+	fmov		s28, s18
+	fmov		s30, s20
+.endm
+
+.macro KERNEL4x8_I
+	ld1	{v0.4s}, [pA], #16
+	ld1	{v4.4s, v5.4s}, [pB], #32
+
+	ldr	d2, [pA], #8
+	ldr	d6, [pB], #8
+	ldr	d7, [pB, #8]
+	ldr	x21, [pA], #8
+	fmul	v16.4s, v0.4s, v4.s[0]
+	ldr	x26, [pB], #16
+	fmul	v18.4s, v0.4s, v4.s[1]
+	ldr	x27, [pB], #8
+	fmul	v20.4s, v0.4s, v4.s[2]
+	fmul	v22.4s, v0.4s, v4.s[3]
+	fmul	v24.4s, v0.4s, v5.s[0]
+	fmul	v26.4s, v0.4s, v5.s[1]
+	fmul	v28.4s, v0.4s, v5.s[2]
+	fmul	v30.4s, v0.4s, v5.s[3]
+.endm
+
+.macro KERNEL4x8_M1
+	ldr	d2, [pA], #8
+	fmov	v0.d[1], x20
+	ldr	d6, [pB], #8
+	fmov	v4.d[1], x24
+	ldr	d7, [pB, #8]
+	fmov	v5.d[1], x25
+	fmla	v16.4s, v0.4s, v4.s[0]
+	ldr	x21, [pA], #8
+	fmla	v18.4s, v0.4s, v4.s[1]
+	ldr	x26, [pB], #16
+	fmla	v20.4s, v0.4s, v4.s[2]
+	ldr	x27, [pB], #8
+	fmla	v22.4s, v0.4s, v4.s[3]
+	fmla	v24.4s, v0.4s, v5.s[0]
+	fmla	v26.4s, v0.4s, v5.s[1]
+	fmla	v28.4s, v0.4s, v5.s[2]
+	fmla	v30.4s, v0.4s, v5.s[3]
+.endm
+
+.macro KERNEL4x8_M2
+	ldr	d0, [pA], #8
+	fmov	v2.d[1], x21
+	ldr	d4, [pB], #8
+	fmov	v6.d[1], x26
+	ldr	d5, [pB, #8]
+	fmov	v7.d[1], x27
+	fmla	v16.4s, v2.4s, v6.s[0]
+	ldr	x20, [pA], #8
+	fmla	v18.4s, v2.4s, v6.s[1]
+	ldr	x24, [pB], #16
+	fmla	v20.4s, v2.4s, v6.s[2]
+	ldr	x25, [pB], #8
+	fmla	v22.4s, v2.4s, v6.s[3]
+	fmla	v24.4s, v2.4s, v7.s[0]
+	fmla	v26.4s, v2.4s, v7.s[1]
+	fmla	v28.4s, v2.4s, v7.s[2]
+	fmla	v30.4s, v2.4s, v7.s[3]
+.endm
+
+.macro KERNEL4x8_E
+	fmov	v2.d[1], x21
+	fmov	v6.d[1], x26
+	fmov	v7.d[1], x27
+	fmla	v16.4s, v2.4s, v6.s[0]
+	fmla	v18.4s, v2.4s, v6.s[1]
+	fmla	v20.4s, v2.4s, v6.s[2]
+	fmla	v22.4s, v2.4s, v6.s[3]
+	fmla	v24.4s, v2.4s, v7.s[0]
+	fmla	v26.4s, v2.4s, v7.s[1]
+	fmla	v28.4s, v2.4s, v7.s[2]
+	fmla	v30.4s, v2.4s, v7.s[3]
+.endm
+
+.macro KERNEL4x8_SUB
+	ld1	{v4.4s}, [pB]
+	add	pB, pB, #16
+	ld1	{v5.4s}, [pB]
+	add	pB, pB, #16
+	ld1	{v0.4s}, [pA]
+	add	pA, pA, #16
+
+	fmla	v16.4s, v0.4s, v4.s[0]
+	fmla	v18.4s, v0.4s, v4.s[1]
+	fmla	v20.4s, v0.4s, v4.s[2]
+	fmla	v22.4s, v0.4s, v4.s[3]
+	fmla	v24.4s, v0.4s, v5.s[0]
+	fmla	v26.4s, v0.4s, v5.s[1]
+	fmla	v28.4s, v0.4s, v5.s[2]
+	fmla	v30.4s, v0.4s, v5.s[3]
+.endm
+
+.macro SAVE4x8
+	add	pCRow1, pCRow0, LDC
+
+
+	fmul	v0.4s, v16.4s, alphaV0
+	st1 	{v0.4s}, [pCRow0]
+
+	add	pCRow2, pCRow1, LDC
+
+
+	fmul	v2.4s, v18.4s, alphaV2
+	st1 	{v2.4s}, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+
+	fmul	v4.4s, v20.4s, alphaV0
+	st1 	{v4.4s}, [pCRow2]
+
+	add	pCRow2, pCRow1, LDC
+
+
+	fmul	v6.4s, v22.4s, alphaV2
+	st1 	{v6.4s}, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+
+	fmul	v0.4s, v24.4s, alphaV0
+	st1 	{v0.4s}, [pCRow2]
+
+	add	pCRow2, pCRow1, LDC
+
+
+	fmul	v2.4s, v26.4s, alphaV2
+	st1 	{v2.4s}, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+
+	fmul	v4.4s, v28.4s, alphaV0
+	st1 	{v4.4s}, [pCRow2]
+
+
+	fmul	v6.4s, v30.4s, alphaV2
+	st1 	{v6.4s}, [pCRow1]
+
+	add	pCRow0, pCRow0, #16
+.endm
+
+/******************************************************************************/
+
+.macro INIT2x8
+	fmov		s16, wzr
+	fmov		s18, wzr
+	fmov		s20, wzr
+	fmov		s22, s16
+	fmov		s24, wzr
+	fmov		s26, s16
+	fmov		s28, s18
+	fmov		s30, s20
+.endm
+
+.macro KERNEL2x8_SUB
+	ld1	{v4.4s}, [pB]
+	add	pB, pB, #16
+	ld1	{v5.4s}, [pB]
+	add	pB, pB, #16
+	ld1	{v0.2s}, [pA]
+	add	pA, pA, #8
+
+	fmla	v16.2s, v0.2s, v4.s[0]
+	fmla	v18.2s, v0.2s, v4.s[1]
+	fmla	v20.2s, v0.2s, v4.s[2]
+	fmla	v22.2s, v0.2s, v4.s[3]
+	fmla	v24.2s, v0.2s, v5.s[0]
+	fmla	v26.2s, v0.2s, v5.s[1]
+	fmla	v28.2s, v0.2s, v5.s[2]
+	fmla	v30.2s, v0.2s, v5.s[3]
+.endm
+
+.macro SAVE2x8
+	add	pCRow1, pCRow0, LDC
+
+
+	fmul	v0.2s, v16.2s, alphaV0
+	st1 	{v0.2s}, [pCRow0]
+
+	add	pCRow2, pCRow1, LDC
+
+
+	fmul	v2.2s, v18.2s, alphaV2
+	st1 	{v2.2s}, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+
+	fmul	v4.2s, v20.2s, alphaV0
+	st1 	{v4.2s}, [pCRow2]
+
+	add	pCRow2, pCRow1, LDC
+
+
+	fmul	v6.2s, v22.2s, alphaV2
+	st1 	{v6.2s}, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+
+	fmul	v0.2s, v24.2s, alphaV0
+	st1 	{v0.2s}, [pCRow2]
+
+	add	pCRow2, pCRow1, LDC
+
+
+	fmul	v2.2s, v26.2s, alphaV2
+	st1 	{v2.2s}, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+
+	fmul	v4.2s, v28.2s, alphaV0
+	st1 	{v4.2s}, [pCRow2]
+
+
+	fmul	v6.2s, v30.2s, alphaV2
+	st1 	{v6.2s}, [pCRow1]
+
+	add	pCRow0, pCRow0, #8
+.endm
+
+/******************************************************************************/
+
+.macro INIT1x8
+	fmov		s16, wzr
+	fmov		s18, wzr
+	fmov		s20, wzr
+	fmov		s22, s16
+	fmov		s24, wzr
+	fmov		s26, s16
+	fmov		s28, s18
+	fmov		s30, s20
+.endm
+
+.macro KERNEL1x8_SUB
+	ld1	{v4.4s}, [pB]
+	add	pB, pB, #16
+	ld1	{v5.4s}, [pB]
+	add	pB, pB, #16
+	ldr	s0, [pA]
+	add	pA, pA, #4
+
+	fmla	s16, s0, v4.s[0]
+	fmla	s18, s0, v4.s[1]
+	fmla	s20, s0, v4.s[2]
+	fmla	s22, s0, v4.s[3]
+	fmla	s24, s0, v5.s[0]
+	fmla	s26, s0, v5.s[1]
+	fmla	s28, s0, v5.s[2]
+	fmla	s30, s0, v5.s[3]
+.endm
+
+.macro SAVE1x8
+	add	pCRow1, pCRow0, LDC
+
+
+	fmul	s0, s16, alphaV0
+	str 	s0, [pCRow0]
+
+	add	pCRow2, pCRow1, LDC
+
+
+	fmul	s2, s18, alphaV2
+	str 	s2, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+
+	fmul	s4, s20, alphaV0
+	str 	s4, [pCRow2]
+
+	add	pCRow2, pCRow1, LDC
+
+
+	fmul	s6, s22, alphaV2
+	str 	s6, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+
+	fmul	s0, s24, alphaV0
+	str 	s0, [pCRow2]
+
+	add	pCRow2, pCRow1, LDC
+
+
+	fmul	s2, s26, alphaV2
+	str 	s2, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+
+	fmul	s4, s28, alphaV0
+	str 	s4, [pCRow2]
+
+
+	fmul	s6, s30, alphaV2
+	str 	s6, [pCRow1]
+
+	add	pCRow0, pCRow0, #4
+.endm
+
+/******************************************************************************/
+
+.macro INIT8x4
+	fmov		s16, wzr
+	fmov		s17, wzr
+	fmov		s20, wzr
+	fmov		s21, s16
+	fmov		s24, wzr
+	fmov		s25, s16
+	fmov		s28, wzr
+	fmov		s29, s16
+.endm
+
+.macro KERNEL8x4_I
+	ld1	{v8.4s}, [pB], #16
+	ld1	{v0.4s, v1.4s}, [pA], #32
+
+	ldr	d9, [pB], #8
+	ldr	d2, [pA], #8
+	ldr	d3, [pA, #8]
+	fmul	v16.4s, v0.4s, v8.s[0]
+	ldr	x25, [pB], #8
+	fmul	v17.4s, v1.4s, v8.s[0]
+	ldr	x22, [pA], #16
+	fmul	v20.4s, v0.4s, v8.s[1]
+	ldr	x23, [pA], #8
+	fmul	v21.4s, v1.4s, v8.s[1]
+	fmul	v24.4s, v0.4s, v8.s[2]
+	fmul	v25.4s, v1.4s, v8.s[2]
+	fmul	v28.4s, v0.4s, v8.s[3]
+	fmul	v29.4s, v1.4s, v8.s[3]
+.endm
+
+.macro KERNEL8x4_M1
+	ldr	d9, [pB], #8
+	fmov	v8.d[1], x24
+	ldr	d2, [pA], #8
+	fmov	v0.d[1], x20
+	ldr	d3, [pA, #8]
+	fmov	v1.d[1], x21
+	fmla	v16.4s, v0.4s, v8.s[0]
+	ldr	x25, [pB], #8
+	fmla	v17.4s, v1.4s, v8.s[0]
+	ldr	x22, [pA], #16
+	fmla	v20.4s, v0.4s, v8.s[1]
+	ldr	x23, [pA], #8
+	fmla	v21.4s, v1.4s, v8.s[1]
+	fmla	v24.4s, v0.4s, v8.s[2]
+	fmla	v25.4s, v1.4s, v8.s[2]
+	fmla	v28.4s, v0.4s, v8.s[3]
+	fmla	v29.4s, v1.4s, v8.s[3]
+.endm
+
+.macro KERNEL8x4_M2
+	ldr	d8, [pB], #8
+	fmov	v9.d[1], x25
+	ldr	d0, [pA], #8
+	fmov	v2.d[1], x22
+	ldr	d1, [pA, #8]
+	fmov	v3.d[1], x23
+	fmla	v16.4s, v2.4s, v9.s[0]
+	ldr	x24, [pB], #8
+	fmla	v17.4s, v3.4s, v9.s[0]
+	ldr	x20, [pA], #16
+	fmla	v20.4s, v2.4s, v9.s[1]
+	ldr	x21, [pA], #8
+	fmla	v21.4s, v3.4s, v9.s[1]
+	fmla	v24.4s, v2.4s, v9.s[2]
+	fmla	v25.4s, v3.4s, v9.s[2]
+	fmla	v28.4s, v2.4s, v9.s[3]
+	fmla	v29.4s, v3.4s, v9.s[3]
+.endm
+
+.macro KERNEL8x4_E
+	fmov	v9.d[1], x25
+	fmov	v2.d[1], x22
+	fmov	v3.d[1], x23
+	fmla	v16.4s, v2.4s, v9.s[0]
+	fmla	v17.4s, v3.4s, v9.s[0]
+	fmla	v20.4s, v2.4s, v9.s[1]
+	fmla	v21.4s, v3.4s, v9.s[1]
+	fmla	v24.4s, v2.4s, v9.s[2]
+	fmla	v25.4s, v3.4s, v9.s[2]
+	fmla	v28.4s, v2.4s, v9.s[3]
+	fmla	v29.4s, v3.4s, v9.s[3]
+.endm
+
+.macro KERNEL8x4_SUB
+	ld1	{v8.4s}, [pB], #16
+	ld1	{v0.4s, v1.4s}, [pA], #32
+	fmla	v16.4s, v0.4s, v8.s[0]
+	fmla	v17.4s, v1.4s, v8.s[0]
+	fmla	v20.4s, v0.4s, v8.s[1]
+	fmla	v21.4s, v1.4s, v8.s[1]
+	fmla	v24.4s, v0.4s, v8.s[2]
+	fmla	v25.4s, v1.4s, v8.s[2]
+	fmla	v28.4s, v0.4s, v8.s[3]
+	fmla	v29.4s, v1.4s, v8.s[3]
+.endm
+
+.macro SAVE8x4
+	add	pCRow1, pCRow0, LDC
+
+
+	fmul	v0.4s, v16.4s, alphaV0
+	fmul	v1.4s, v17.4s, alphaV1
+	st1 	{v0.4s, v1.4s}, [pCRow0]
+
+	add	pCRow2, pCRow1, LDC
+
+
+	fmul	v4.4s, v20.4s, alphaV0
+	fmul	v5.4s, v21.4s, alphaV1
+	st1 	{v4.4s, v5.4s}, [pCRow1]
+
+	add	pCRow1, pCRow2, LDC
+
+
+	fmul	v0.4s, v24.4s, alphaV0
+	fmul	v1.4s, v25.4s, alphaV1
+	st1 	{v0.4s, v1.4s}, [pCRow2]
+
+
+	fmul	v4.4s, v28.4s, alphaV0
+	fmul	v5.4s, v29.4s, alphaV1
+	st1 	{v4.4s, v5.4s}, [pCRow1]
+
+	add	pCRow0, pCRow0, #32
+.endm
+
+/******************************************************************************/
+
+
+.macro INIT4x4
+	fmov		s16, wzr
+	fmov		s17, s16
+	fmov		s20, s17
+	fmov		s21, s16
+	fmov		s24, s17
+	fmov		s25, s16
+	fmov		s28, s17
+	fmov		s29, s16
+.endm
+
+.macro KERNEL4x4_I
+	ld1	{v8.2s, v9.2s}, [pB]
+	add	pB, pB, #16
+	ld1	{v0.2s, v1.2s}, [pA]
+	add	pA, pA, #16
+
+	fmul	v16.2s, v0.2s, v8.s[0]
+	fmul	v29.2s, v1.2s, v9.s[1]
+
+	fmul	v20.2s, v0.2s, v8.s[1]
+	fmul	v25.2s, v1.2s, v9.s[0]
+
+	fmul	v24.2s, v0.2s, v9.s[0]
+	fmul	v21.2s, v1.2s, v8.s[1]
+
+	fmul	v28.2s, v0.2s, v9.s[1]
+	fmul	v17.2s, v1.2s, v8.s[0]
+
+	ld1	{v12.2s, v13.2s}, [pB]
+	add	pB, pB, #16
+	ld1	{v4.2s, v5.2s}, [pA]
+	add	pA, pA, #16
+.endm
+
+.macro KERNEL4x4_M1
+	fmla	v16.2s, v0.2s, v8.s[0]
+	fmla	v29.2s, v1.2s, v9.s[1]
+
+	ld1	{v12.2s, v13.2s}, [pB]		// For next round
+	add	pB, pB, #16
+
+	fmla	v20.2s, v0.2s, v8.s[1]
+	fmla	v25.2s, v1.2s, v9.s[0]
+
+	ld1	{v4.2s, v5.2s}, [pA]		// For next round
+	add	pA, pA, #16
+
+	fmla	v24.2s, v0.2s, v9.s[0]
+	fmla	v21.2s, v1.2s, v8.s[1]
+
+	prfm	PLDL1KEEP, [pB, #512]
+
+	fmla	v28.2s, v0.2s, v9.s[1]
+	fmla	v17.2s, v1.2s, v8.s[0]
+.endm
+
+.macro KERNEL4x4_M2
+	fmla	v16.2s, v4.2s, v12.s[0]
+	fmla	v29.2s, v5.2s, v13.s[1]
+
+	ld1	{v8.2s, v9.2s}, [pB]		// For next round
+	add	pB, pB, #16
+
+	fmla	v20.2s, v4.2s, v12.s[1]
+	fmla	v25.2s, v5.2s, v13.s[0]
+
+	ld1	{v0.2s, v1.2s}, [pA]		// For next round
+	add	pA, pA, #16
+
+	fmla	v24.2s, v4.2s, v13.s[0]
+	fmla	v21.2s, v5.2s, v12.s[1]
+
+	prfm	PLDL1KEEP, [pA, #512]
+
+	fmla	v28.2s, v4.2s, v13.s[1]
+	fmla	v17.2s, v5.2s, v12.s[0]
+.endm
+
+.macro KERNEL4x4_E
+	fmla	v16.2s, v4.2s, v12.s[0]
+	fmla	v29.2s, v5.2s, v13.s[1]
+
+	fmla	v20.2s, v4.2s, v12.s[1]
+	fmla	v25.2s, v5.2s, v13.s[0]
+
+	fmla	v24.2s, v4.2s, v13.s[0]
+	fmla	v21.2s, v5.2s, v12.s[1]
+
+	fmla	v28.2s, v4.2s, v13.s[1]
+	fmla	v17.2s, v5.2s, v12.s[0]
+.endm
+
+.macro KERNEL4x4_SUB
+	ld1	{v8.2s, v9.2s}, [pB]
+	add	pB, pB, #16
+	ld1	{v0.2s, v1.2s}, [pA]
+	add	pA, pA, #16
+
+	fmla	v16.2s, v0.2s, v8.s[0]
+	fmla	v29.2s, v1.2s, v9.s[1]
+
+	fmla	v20.2s, v0.2s, v8.s[1]
+	fmla	v25.2s, v1.2s, v9.s[0]
+
+	fmla	v24.2s, v0.2s, v9.s[0]
+	fmla	v21.2s, v1.2s, v8.s[1]
+
+	fmla	v28.2s, v0.2s, v9.s[1]
+	fmla	v17.2s, v1.2s, v8.s[0]
+.endm
+
+.macro SAVE4x4
+
+	fmul	v8.2s, v16.2s, alphaV0
+	fmul	v9.2s, v17.2s, alphaV1
+	st1 	{v8.2s, v9.2s}, [pCRow0]
+
+	add	pCRow1, pCRow0, LDC
+
+	fmul	v12.2s, v20.2s, alphaV2
+	fmul	v13.2s, v21.2s, alphaV3
+	st1 	{v12.2s, v13.2s}, [pCRow1]
+
+	add	pCRow2, pCRow1, LDC
+
+	fmul	v8.2s, v24.2s, alphaV0
+	fmul	v9.2s, v25.2s, alphaV1
+	st1 	{v8.2s, v9.2s}, [pCRow2]
+
+	add	pCRow1, pCRow2, LDC
+
+	fmul	v12.2s, v28.2s, alphaV2
+	fmul	v13.2s, v29.2s, alphaV3
+	st1 	{v12.2s, v13.2s}, [pCRow1]
+
+	add	pCRow0, pCRow0, #16
+.endm
+
+/******************************************************************************/
+
+.macro INIT2x4
+	fmov		s16, wzr
+	fmov		s20, s16
+	fmov		s24, s20
+	fmov		s28, s16
+.endm
+
+.macro KERNEL2x4_SUB
+	ld1	{v8.2s, v9.2s}, [pB]
+	add	pB, pB, #16
+	ld1	{v0.2s}, [pA]
+	add	pA, pA, #8
+
+	fmla	v16.2s, v0.2s, v8.s[0]
+	fmla	v20.2s, v0.2s, v8.s[1]
+	fmla	v24.2s, v0.2s, v9.s[0]
+	fmla	v28.2s, v0.2s, v9.s[1]
+.endm
+
+.macro SAVE2x4
+
+	fmul	v8.2s, v16.2s, alphaV0
+	st1	{v8.2s}, [pCRow0]
+
+	add	pCRow1, pCRow0, LDC
+
+	fmul	v12.2s, v20.2s, alphaV1
+	st1	{v12.2s}, [pCRow1]
+
+	add	pCRow2, pCRow1, LDC
+
+	fmul	v8.2s, v24.2s, alphaV2
+	st1	{v8.2s}, [pCRow2]
+
+	add	pCRow1, pCRow2, LDC
+
+	fmul	v12.2s, v28.2s, alphaV3
+	st1	{v12.2s}, [pCRow1]
+
+	add	pCRow0, pCRow0, #8
+.endm
+
+/******************************************************************************/
+
+.macro INIT1x4
+	fmov		s16, wzr
+	fmov		s20, s16
+.endm
+
+.macro KERNEL1x4_SUB
+	ldr	s0, [pA]
+	add	pA, pA, #4
+
+	ld1	{v8.2s, v9.2s}, [pB]
+	add	pB, pB, #16
+
+	fmla	v16.2s, v8.2s, v0.s[0]
+	fmla	v20.2s, v9.2s, v0.s[0]
+.endm
+
+.macro SAVE1x4
+	add	pCRow1, pCRow0, LDC
+
+
+	fmul	v8.2s, v16.2s, alphaV0
+	st1	{v8.s}[0], [pCRow0]
+	st1	{v8.s}[1], [pCRow1]
+
+	add	pCRow2, pCRow1, LDC
+	add	pCRow1, pCRow2, LDC
+
+
+	fmul	v12.2s, v20.2s, alphaV1
+	st1	{v12.s}[0], [pCRow2]
+	st1	{v12.s}[1], [pCRow1]
+
+	add	pCRow0, pCRow0, #4
+.endm
+
+/******************************************************************************/
+
+.macro INIT8x2
+	fmov	s16, wzr
+	fmov	s17, s16
+	fmov	s20, s17
+	fmov	s21, s16
+.endm
+
+.macro KERNEL8x2_SUB
+	ld1	{v8.2s}, [pB]
+	add	pB, pB, #8
+	ld1	{v0.4s}, [pA]
+	add	pA, pA, #16
+	ld1	{v1.4s}, [pA]
+	add	pA, pA, #16
+
+	fmla	v16.4s, v0.4s, v8.s[0]
+	fmla	v17.4s, v1.4s, v8.s[0]
+
+	fmla	v20.4s, v0.4s, v8.s[1]
+	fmla	v21.4s, v1.4s, v8.s[1]
+.endm
+
+.macro SAVE8x2
+	add	pCRow1, pCRow0, LDC
+
+
+	fmul	v0.4s, v16.4s, alphaV0
+	fmul	v1.4s, v17.4s, alphaV1
+	st1 	{v0.4s, v1.4s}, [pCRow0]
+
+	add	pCRow2, pCRow1, LDC
+
+
+	fmul	v4.4s, v20.4s, alphaV0
+	fmul	v5.4s, v21.4s, alphaV1
+	st1 	{v4.4s, v5.4s}, [pCRow1]
+
+	add	pCRow0, pCRow0, #32
+.endm
+
+/******************************************************************************/
+
+.macro INIT4x2
+	fmov	s16, wzr
+	fmov	s17, s16
+	fmov	s20, s17
+	fmov	s21, s16
+.endm
+
+.macro KERNEL4x2_SUB
+	ld1	{v8.2s}, [pB]
+	add	pB, pB, #8
+	ld1	{v0.2s, v1.2s}, [pA]
+	add	pA, pA, #16
+
+	fmla	v16.2s, v0.2s, v8.s[0]
+	fmla	v17.2s, v1.2s, v8.s[0]
+	fmla	v20.2s, v0.2s, v8.s[1]
+	fmla	v21.2s, v1.2s, v8.s[1]
+.endm
+
+.macro SAVE4x2
+
+	fmul	v8.2s, v16.2s, alphaV0
+	fmul	v9.2s, v17.2s, alphaV1
+	st1	{v8.2s, v9.2s}, [pCRow0]
+
+	add	pCRow1, pCRow0, LDC
+
+	fmul	v12.2s, v20.2s, alphaV2
+	fmul	v13.2s, v21.2s, alphaV3
+	st1	{v12.2s, v13.2s}, [pCRow1]
+
+	add	pCRow0, pCRow0, #16
+.endm
+
+/******************************************************************************/
+
+.macro INIT2x2
+	fmov		s16, wzr
+	fmov		s20, s16
+.endm
+
+.macro KERNEL2x2_SUB
+	ld1	{v8.2s}, [pB]
+	add	pB, pB, #8
+
+	ld1	{v0.2s}, [pA]
+	add	pA, pA, #8
+
+	fmla	v16.2s, v0.2s, v8.s[0]
+	fmla	v20.2s, v0.2s, v8.s[1]
+.endm
+
+.macro SAVE2x2
+
+	fmul	v8.2s, v16.2s, alphaV0
+	st1	{v8.2s}, [pCRow0]
+
+	add	pCRow1 , pCRow0, LDC
+
+	fmul	v12.2s, v20.2s, alphaV1
+	st1	{v12.2s}, [pCRow1]
+
+	add	pCRow0, pCRow0, #8
+.endm
+
+/******************************************************************************/
+
+.macro INIT1x2
+	fmov		s16, wzr
+.endm
+
+.macro KERNEL1x2_SUB
+	ld1	{v8.2s} , [pB]
+	add	pB , pB, #8
+
+	ldr	s0 , [pA]
+	add	pA, pA, #4
+
+	fmla	v16.2s, v8.2s, v0.s[0]
+.endm
+
+.macro SAVE1x2
+	add	pCRow1 , pCRow0, LDC
+
+
+	fmul	v8.2s, v16.2s, alphaV0
+	st1	{v8.s}[0], [pCRow0]
+	st1	{v8.s}[1], [pCRow1]
+
+	add	pCRow0, pCRow0, #4
+.endm
+
+/******************************************************************************/
+
+.macro INIT8x1
+	fmov	s16, wzr
+	fmov	s17, wzr
+.endm
+
+.macro KERNEL8x1_SUB
+	ldr	s8, [pB]
+	add	pB , pB, #4
+
+	ld1	{v0.4s}, [pA]
+	add	pA, pA, #16
+	ld1	{v1.4s}, [pA]
+	add	pA, pA, #16
+
+	fmla	v16.4s, v0.4s, v8.s[0]
+	fmla	v17.4s, v1.4s, v8.s[0]
+.endm
+
+.macro SAVE8x1
+
+	fmul	v0.4s, v16.4s, alphaV0
+	fmul	v1.4s, v17.4s, alphaV1
+	st1 	{v0.4s, v1.4s}, [pCRow0]
+
+	add	pCRow0, pCRow0, #32
+.endm
+
+/******************************************************************************/
+
+.macro INIT4x1
+	fmov	s16, wzr
+	fmov	s17, s16
+.endm
+
+.macro KERNEL4x1_SUB
+	ldr	s8, [pB]
+	add	pB , pB, #4
+
+	ld1	{v0.2s, v1.2s}, [pA]
+	add	pA , pA, #16
+
+	fmla	v16.2s, v0.2s, v8.s[0]
+	fmla	v17.2s, v1.2s, v8.s[0]
+.endm
+
+.macro SAVE4x1
+
+	fmul	v8.2s, v16.2s, alphaV0
+	fmul	v9.2s, v17.2s, alphaV1
+	st1	{v8.2s, v9.2s}, [pCRow0]
+
+	add	pCRow0, pCRow0, #16
+.endm
+
+/******************************************************************************/
+
+.macro INIT2x1
+	fmov		s16, wzr
+.endm
+
+.macro KERNEL2x1_SUB
+	ldr	s8, [pB]
+	add	pB , pB, #4
+
+	ld1	{v0.2s}, [pA]
+	add	pA , pA, #8
+
+	fmla	v16.2s, v0.2s, v8.s[0]
+.endm
+
+.macro SAVE2x1
+
+	fmul	v8.2s, v16.2s, alphaV0
+	st1	{v8.2s}, [pCRow0]
+
+	add	pCRow0, pCRow0, #8
+.endm
+
+/******************************************************************************/
+
+.macro INIT1x1
+	fmov	s16, wzr
+.endm
+
+.macro KERNEL1x1_SUB
+	ldr	s8, [pB]
+	add	pB , pB, #4
+
+	ldr	s0, [pA]
+	add	pA , pA, #4
+
+	fmadd 	s16, s0, s8, s16  
+.endm
+
+.macro SAVE1x1
+
+	fmul	s8, s16, alpha0
+	str 	s8, [pCRow0]
+
+	add	pCRow0, pCRow0, #4
+.endm
+
+/*******************************************************************************
+* End of macro definitions
+*******************************************************************************/
+
+	PROLOGUE
+
+.Lstrmm_kernel_begin:
+
+	.align 5
+	add	sp, sp, #-(11 * 16)
+	stp	d8, d9, [sp, #(0 * 16)]
+	stp	d10, d11, [sp, #(1 * 16)]
+	stp	d12, d13, [sp, #(2 * 16)]
+	stp	d14, d15, [sp, #(3 * 16)]
+	stp	d16, d17, [sp, #(4 * 16)]
+	stp	x18, x19, [sp, #(5 * 16)]
+	stp	x20, x21, [sp, #(6 * 16)]
+	stp	x22, x23, [sp, #(7 * 16)]
+	stp	x24, x25, [sp, #(8 * 16)]
+	stp	x26, x27, [sp, #(9 * 16)]
+	str	x28, [sp, #(10 * 16)]
+
+	fmov	alpha0, s0
+	fmov	alpha1, s0
+	fmov	alpha2, s0
+	fmov	alpha3, s0
+
+	lsl	LDC, LDC, #2			// ldc = ldc * 4
+
+#if !defined(LEFT)
+	neg	tempOffset, offset
+#endif
+	mov	pB, origPB
+
+	mov	counterJ, origN
+	asr 	counterJ, counterJ, #3		// J = J / 8
+	cmp 	counterJ, #0
+	ble	.Lstrmm_kernel_L4_BEGIN
+
+/******************************************************************************/
+/******************************************************************************/
+
+.Lstrmm_kernel_L8_BEGIN:
+	mov	pCRow0, pC			// pCRow0 = C
+	add	pC, pC, LDC, lsl #3
+
+#if defined(LEFT)
+	mov	tempOffset, offset
+#endif
+
+	mov	pA, origPA			// pA = start of A array
+
+/******************************************************************************/
+
+.Lstrmm_kernel_L8_M8_BEGIN:
+
+	mov	counterI, origM
+	asr 	counterI, counterI, #3		// counterI = counterI / 8
+	cmp 	counterI, #0
+	ble	.Lstrmm_kernel_L8_M4_BEGIN
+
+.Lstrmm_kernel_L8_M8_20:
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	mov	pB, origPB
+#else
+	mov	pB, origPB
+	lsl	temp, tempOffset, #5
+	add	pA, pA, temp
+	add	pB, pB, temp
+#endif
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#elif defined(LEFT)
+	add	tempK, tempOffset, #8
+#else
+	add	tempK, tempOffset, #8
+#endif
+
+	asr 	counterL , tempK, #3		// L = K / 8
+	cmp	counterL , #2			// is there at least 16 to do?
+	blt	.Lstrmm_kernel_L8_M8_32
+
+	KERNEL8x8_I				// do one in the K
+	KERNEL8x8_M2				// do another in the K
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+
+	subs	counterL, counterL, #2
+	ble	.Lstrmm_kernel_L8_M8_22a
+	.align 5
+
+.Lstrmm_kernel_L8_M8_22:
+
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L8_M8_22
+
+.Lstrmm_kernel_L8_M8_22a:
+
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_E
+
+	b	 .Lstrmm_kernel_L8_M8_44
+
+.Lstrmm_kernel_L8_M8_32:
+
+	tst	counterL, #1
+	ble	.Lstrmm_kernel_L8_M8_40
+
+	KERNEL8x8_I
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_M2
+	KERNEL8x8_M1
+	KERNEL8x8_E
+
+	b	.Lstrmm_kernel_L8_M8_44
+
+.Lstrmm_kernel_L8_M8_40:
+
+	INIT8x8
+
+.Lstrmm_kernel_L8_M8_44:
+
+	ands	counterL , tempK, #7
+	ble	.Lstrmm_kernel_L8_M8_100
+
+.Lstrmm_kernel_L8_M8_46:
+
+	KERNEL8x8_SUB
+
+	subs	counterL, counterL, 1
+	bgt	.Lstrmm_kernel_L8_M8_46
+
+.Lstrmm_kernel_L8_M8_100:
+
+	SAVE8x8
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#if defined(LEFT)
+	sub	tempK, tempK, #8
+#else
+	sub	tempK, tempK, #8
+#endif
+	lsl	temp, tempK, #5
+	add	pA, pA, temp
+	add	pB, pB, temp
+#endif
+#if defined(LEFT)
+	add	tempOffset, tempOffset, #8
+#endif
+
+.Lstrmm_kernel_L8_M8_END:
+	subs	counterI, counterI, #1
+	bne	.Lstrmm_kernel_L8_M8_20
+
+/******************************************************************************/
+
+.Lstrmm_kernel_L8_M4_BEGIN:
+
+	mov	counterI, origM
+	tst	counterI , #7
+	ble	.Lstrmm_kernel_L8_END
+
+	tst	counterI, #4
+	ble	.Lstrmm_kernel_L8_M2_BEGIN
+
+.Lstrmm_kernel_L8_M4_20:
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	mov	pB, origPB
+#else
+	mov	pB, origPB
+	lsl	temp, tempOffset, #4
+	add	pA, pA, temp
+	lsl	temp, tempOffset, #5
+	add	pB, pB, temp
+#endif
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#elif defined(LEFT)
+	add	tempK, tempOffset, #4
+#else
+	add	tempK, tempOffset, #8
+#endif
+
+	asr 	counterL , tempK, #1		// L = K / 2
+	cmp	counterL , #2			// is there at least 4 to do?
+	blt	.Lstrmm_kernel_L8_M4_32
+
+	KERNEL4x8_I				// do one in the K
+	KERNEL4x8_M2				// do another in the K
+
+	subs	counterL, counterL, #2
+	ble	.Lstrmm_kernel_L8_M4_22a
+	.align 5
+
+.Lstrmm_kernel_L8_M4_22:
+
+	KERNEL4x8_M1
+	KERNEL4x8_M2
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L8_M4_22
+
+.Lstrmm_kernel_L8_M4_22a:
+
+	KERNEL4x8_M1
+	KERNEL4x8_E
+
+	b	 .Lstrmm_kernel_L8_M4_44
+
+.Lstrmm_kernel_L8_M4_32:
+
+	tst	counterL, #1
+	ble	.Lstrmm_kernel_L8_M4_40
+
+	KERNEL4x8_I
+	KERNEL4x8_E
+
+	b	.Lstrmm_kernel_L8_M4_44
+
+.Lstrmm_kernel_L8_M4_40:
+
+	INIT4x8
+
+.Lstrmm_kernel_L8_M4_44:
+
+	ands	counterL , tempK, #1
+	ble	.Lstrmm_kernel_L8_M4_100
+
+.Lstrmm_kernel_L8_M4_46:
+
+	KERNEL4x8_SUB
+
+.Lstrmm_kernel_L8_M4_100:
+
+	SAVE4x8
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#if defined(LEFT)
+	sub	tempK, tempK, #4
+#else
+	sub	tempK, tempK, #8
+#endif
+	lsl	temp, tempK, #4
+	add	pA, pA, temp
+	lsl	temp, tempK, #5
+	add	pB, pB, temp
+#endif
+#if defined(LEFT)
+	add	tempOffset, tempOffset, #4
+#endif
+
+.Lstrmm_kernel_L8_M4_END:
+
+/******************************************************************************/
+
+.Lstrmm_kernel_L8_M2_BEGIN:
+
+	mov	counterI, origM
+	tst	counterI , #3
+	ble	.Lstrmm_kernel_L8_END
+
+	tst	counterI, #2			// counterI = counterI / 2
+	ble	.Lstrmm_kernel_L8_M1_BEGIN
+
+.Lstrmm_kernel_L8_M2_20:
+
+	INIT2x8
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	mov	pB, origPB
+#else
+	mov	pB, origPB
+	lsl	temp, tempOffset, #3
+	add	pA, pA, temp
+	lsl	temp, tempOffset, #5
+	add	pB, pB, temp
+#endif
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#elif defined(LEFT)
+	add	tempK, tempOffset, #2
+#else
+	add	tempK, tempOffset, #8
+#endif
+
+	asr 	counterL , tempK, #3		// counterL = counterL / 8
+	cmp	counterL , #0
+	ble	.Lstrmm_kernel_L8_M2_40
+
+.Lstrmm_kernel_L8_M2_22:
+
+	KERNEL2x8_SUB
+	KERNEL2x8_SUB
+	KERNEL2x8_SUB
+	KERNEL2x8_SUB
+
+	KERNEL2x8_SUB
+	KERNEL2x8_SUB
+	KERNEL2x8_SUB
+	KERNEL2x8_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L8_M2_22
+
+
+.Lstrmm_kernel_L8_M2_40:
+
+	ands	counterL , tempK, #7		// counterL = counterL % 8
+	ble	.Lstrmm_kernel_L8_M2_100
+
+.Lstrmm_kernel_L8_M2_42:
+
+	KERNEL2x8_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L8_M2_42
+
+.Lstrmm_kernel_L8_M2_100:
+
+	SAVE2x8
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#if defined(LEFT)
+	sub	tempK, tempK, #2
+#else
+	sub	tempK, tempK, #8
+#endif
+	lsl	temp, tempK, #3
+	add	pA, pA, temp
+	lsl	temp, tempK, #5
+	add	pB, pB, temp
+#endif
+#if defined(LEFT)
+	add	tempOffset, tempOffset, #2
+#endif
+
+.Lstrmm_kernel_L8_M2_END:
+
+/******************************************************************************/
+
+.Lstrmm_kernel_L8_M1_BEGIN:
+
+	tst	counterI, #1			// counterI = counterI % 2
+	ble	.Lstrmm_kernel_L8_END
+
+.Lstrmm_kernel_L8_M1_20:
+
+	INIT1x8
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	mov	pB, origPB
+#else
+	mov	pB, origPB
+	lsl	temp, tempOffset, #2
+	add	pA, pA, temp
+	lsl	temp, tempOffset, #5
+	add	pB, pB, temp
+#endif
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#elif defined(LEFT)
+	add	tempK, tempOffset, #1
+#else
+	add	tempK, tempOffset, #8
+#endif
+
+	asr 	counterL , tempK, #3		// counterL = counterL / 8
+	cmp	counterL , #0
+	ble	.Lstrmm_kernel_L8_M1_40
+
+.Lstrmm_kernel_L8_M1_22:
+	KERNEL1x8_SUB
+	KERNEL1x8_SUB
+	KERNEL1x8_SUB
+	KERNEL1x8_SUB
+
+	KERNEL1x8_SUB
+	KERNEL1x8_SUB
+	KERNEL1x8_SUB
+	KERNEL1x8_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L8_M1_22
+
+
+.Lstrmm_kernel_L8_M1_40:
+
+	ands	counterL , tempK, #7		// counterL = counterL % 8
+	ble	.Lstrmm_kernel_L8_M1_100
+
+.Lstrmm_kernel_L8_M1_42:
+
+	KERNEL1x8_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L8_M1_42
+
+.Lstrmm_kernel_L8_M1_100:
+
+	SAVE1x8
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#if defined(LEFT)
+	sub	tempK, tempK, #1
+#else
+	sub	tempK, tempK, #8
+#endif
+	lsl	temp, tempK, #2
+	add	pA, pA, temp
+	lsl	temp, tempK, #5
+	add	pB, pB, temp
+#endif
+#if defined(LEFT)
+	add	tempOffset, tempOffset, #1
+#endif
+
+.Lstrmm_kernel_L8_END:
+	lsl	temp, origK, #5			// B = B + K * 4 * 8
+	add	origPB, origPB, temp
+
+#if !defined(LEFT)
+	add	tempOffset, tempOffset, #8
+#endif
+
+	subs	counterJ, counterJ , #1		// j--
+	bgt	.Lstrmm_kernel_L8_BEGIN
+
+/******************************************************************************/
+/******************************************************************************/
+
+.Lstrmm_kernel_L4_BEGIN:
+
+	mov	counterJ , origN
+	tst	counterJ , #7
+	ble	.Lstrmm_kernel_L999
+
+	tst	counterJ , #4
+	ble	.Lstrmm_kernel_L2_BEGIN
+
+	mov	pCRow0, pC			// pCRow0 = pC
+
+	add	pC,pC,LDC, lsl #2
+
+#if defined(LEFT)
+	mov	tempOffset, offset
+#endif
+
+	mov	pA, origPA			// pA = A
+
+/******************************************************************************/
+
+.Lstrmm_kernel_L4_M8_BEGIN:
+
+	mov	counterI, origM
+	asr 	counterI, counterI, #3		// counterI = counterI / 8
+	cmp 	counterI, #0
+	ble	.Lstrmm_kernel_L4_M4_BEGIN
+
+.Lstrmm_kernel_L4_M8_20:
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	mov	pB, origPB
+#else
+	mov	pB, origPB
+	lsl	temp, tempOffset, #5
+	add	pA, pA, temp
+	lsl	temp, tempOffset, #4
+	add	pB, pB, temp
+#endif
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#elif defined(LEFT)
+	add	tempK, tempOffset, #8
+#else
+	add	tempK, tempOffset, #4
+#endif
+
+	asr 	counterL , tempK, #1		// L = K / 2
+	cmp	counterL , #2			// is there at least 4 to do?
+	blt	.Lstrmm_kernel_L4_M8_32
+
+	KERNEL8x4_I				// do one in the K
+	KERNEL8x4_M2				// do another in the K
+
+	subs	counterL, counterL, #2
+	ble	.Lstrmm_kernel_L4_M8_22a
+	.align 5
+
+.Lstrmm_kernel_L4_M8_22:
+
+	KERNEL8x4_M1
+	KERNEL8x4_M2
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L4_M8_22
+
+.Lstrmm_kernel_L4_M8_22a:
+
+	KERNEL8x4_M1
+	KERNEL8x4_E
+
+	b	 .Lstrmm_kernel_L4_M8_44
+
+.Lstrmm_kernel_L4_M8_32:
+
+	tst	counterL, #1
+	ble	.Lstrmm_kernel_L4_M8_40
+
+	KERNEL8x4_I
+	KERNEL8x4_E
+
+	b	.Lstrmm_kernel_L4_M8_44
+
+.Lstrmm_kernel_L4_M8_40:
+
+	INIT8x4
+
+.Lstrmm_kernel_L4_M8_44:
+
+	ands	counterL , tempK, #1
+	ble	.Lstrmm_kernel_L4_M8_100
+
+.Lstrmm_kernel_L4_M8_46:
+
+	KERNEL8x4_SUB
+
+.Lstrmm_kernel_L4_M8_100:
+
+	SAVE8x4
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#if defined(LEFT)
+	sub	tempK, tempK, #8
+#else
+	sub	tempK, tempK, #4
+#endif
+	lsl	temp, tempK, #5
+	add	pA, pA, temp
+	lsl	temp, tempK, #4
+	add	pB, pB, temp
+#endif
+#if defined(LEFT)
+	add	tempOffset, tempOffset, #8
+#endif
+.Lstrmm_kernel_L4_M8_END:
+	subs	counterI, counterI, #1
+	bne	.Lstrmm_kernel_L4_M8_20
+
+/******************************************************************************/
+
+.Lstrmm_kernel_L4_M4_BEGIN:
+
+	mov	counterI, origM
+	tst	counterI , #7
+	ble	.Lstrmm_kernel_L4_END
+
+	tst	counterI, #4
+	ble	.Lstrmm_kernel_L4_M2_BEGIN
+
+.Lstrmm_kernel_L4_M4_20:
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	mov	pB, origPB
+#else
+	mov	pB, origPB
+	lsl	temp, tempOffset, #4
+	add	pB, pB, temp
+	add	pA, pA, temp
+#endif
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#elif defined(LEFT)
+	add	tempK, tempOffset, #4
+#else
+	add	tempK, tempOffset, #4
+#endif
+	asr 	counterL , tempK, #1		// L = K / 2
+	cmp	counterL , #2			// is there at least 4 to do?
+	blt	.Lstrmm_kernel_L4_M4_32
+
+	KERNEL4x4_I				// do one in the K
+	KERNEL4x4_M2				// do another in the K
+
+	subs	counterL, counterL, #2
+	ble	.Lstrmm_kernel_L4_M4_22a
+	.align 5
+
+.Lstrmm_kernel_L4_M4_22:
+
+	KERNEL4x4_M1
+	KERNEL4x4_M2
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L4_M4_22
+
+.Lstrmm_kernel_L4_M4_22a:
+
+	KERNEL4x4_M1
+	KERNEL4x4_E
+
+	b	 .Lstrmm_kernel_L4_M4_44
+
+.Lstrmm_kernel_L4_M4_32:
+
+	tst	counterL, #1
+	ble	.Lstrmm_kernel_L4_M4_40
+
+	KERNEL4x4_I
+	KERNEL4x4_E
+
+	b	.Lstrmm_kernel_L4_M4_44
+
+.Lstrmm_kernel_L4_M4_40:
+
+	INIT4x4
+
+.Lstrmm_kernel_L4_M4_44:
+
+	ands	counterL , tempK, #1
+	ble	.Lstrmm_kernel_L4_M4_100
+
+.Lstrmm_kernel_L4_M4_46:
+
+	KERNEL4x4_SUB
+
+.Lstrmm_kernel_L4_M4_100:
+
+	SAVE4x4
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#if defined(LEFT)
+	sub	tempK, tempK, #4
+#else
+	sub	tempK, tempK, #4
+#endif
+	lsl	temp, tempK, #4
+	add	pA, pA, temp
+	add	pB, pB, temp
+#endif
+#if defined(LEFT)
+	add	tempOffset, tempOffset, #4
+#endif
+.Lstrmm_kernel_L4_M4_END:
+
+/******************************************************************************/
+
+.Lstrmm_kernel_L4_M2_BEGIN:
+
+	mov	counterI, origM
+	tst	counterI , #3
+	ble	.Lstrmm_kernel_L4_END
+
+	tst	counterI, #2			// counterI = counterI / 2
+	ble	.Lstrmm_kernel_L4_M1_BEGIN
+
+.Lstrmm_kernel_L4_M2_20:
+
+	INIT2x4
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	mov	pB, origPB
+#else
+	mov	pB, origPB
+	lsl	temp, tempOffset, #3
+	add	pA, pA, temp
+	lsl	temp, tempOffset, #4
+	add	pB, pB, temp
+#endif
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#elif defined(LEFT)
+	add	tempK, tempOffset, #2
+#else
+	add	tempK, tempOffset, #4
+#endif
+	asr 	counterL , tempK, #3		// counterL = counterL / 8
+	cmp	counterL , #0
+	ble	.Lstrmm_kernel_L4_M2_40
+
+.Lstrmm_kernel_L4_M2_22:
+
+	KERNEL2x4_SUB
+	KERNEL2x4_SUB
+	KERNEL2x4_SUB
+	KERNEL2x4_SUB
+
+	KERNEL2x4_SUB
+	KERNEL2x4_SUB
+	KERNEL2x4_SUB
+	KERNEL2x4_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L4_M2_22
+
+
+.Lstrmm_kernel_L4_M2_40:
+
+	ands	counterL , tempK, #7		// counterL = counterL % 8
+	ble	.Lstrmm_kernel_L4_M2_100
+
+.Lstrmm_kernel_L4_M2_42:
+
+	KERNEL2x4_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L4_M2_42
+
+.Lstrmm_kernel_L4_M2_100:
+
+	SAVE2x4
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#if defined(LEFT)
+	sub	tempK, tempK, #2
+#else
+	sub	tempK, tempK, #4
+#endif
+	lsl	temp, tempK, #3
+	add	pA, pA, temp
+	lsl	temp, tempK, #4
+	add	pB, pB, temp
+#endif
+#if defined(LEFT)
+	add	tempOffset, tempOffset, #2
+#endif
+.Lstrmm_kernel_L4_M2_END:
+
+/******************************************************************************/
+
+.Lstrmm_kernel_L4_M1_BEGIN:
+
+	tst	counterI, #1			// counterI = counterI % 2
+	ble	.Lstrmm_kernel_L4_END
+
+.Lstrmm_kernel_L4_M1_20:
+
+	INIT1x4
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	mov	pB, origPB
+#else
+	mov	pB, origPB
+	lsl	temp, tempOffset, #4
+	add	pB, pB, temp
+	lsl	temp, tempOffset, #2
+	add	pA, pA, temp
+#endif
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#elif defined(LEFT)
+	add	tempK, tempOffset, #1
+#else
+	add	tempK, tempOffset, #4
+#endif
+	asr 	counterL , tempK, #3		// counterL = counterL / 8
+	cmp	counterL , #0
+	ble	.Lstrmm_kernel_L4_M1_40
+
+.Lstrmm_kernel_L4_M1_22:
+	KERNEL1x4_SUB
+	KERNEL1x4_SUB
+	KERNEL1x4_SUB
+	KERNEL1x4_SUB
+
+	KERNEL1x4_SUB
+	KERNEL1x4_SUB
+	KERNEL1x4_SUB
+	KERNEL1x4_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L4_M1_22
+
+
+.Lstrmm_kernel_L4_M1_40:
+
+	ands	counterL , tempK, #7		// counterL = counterL % 8
+	ble	.Lstrmm_kernel_L4_M1_100
+
+.Lstrmm_kernel_L4_M1_42:
+
+	KERNEL1x4_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L4_M1_42
+
+.Lstrmm_kernel_L4_M1_100:
+
+	SAVE1x4
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#if defined(LEFT)
+	sub	tempK, tempK, #1
+#else
+	sub	tempK, tempK, #4
+#endif
+	lsl	temp, tempK, #2
+	add	pA, pA, temp
+	lsl	temp, tempK, #4
+	add	pB, pB, temp
+#endif
+#if defined(LEFT)
+	add	tempOffset, tempOffset, #1
+#endif
+.Lstrmm_kernel_L4_END:
+	add	origPB, origPB, origK, lsl #4	// B = B + K * 4 * 4
+#if !defined(LEFT)
+	add	tempOffset, tempOffset, #4
+#endif
+
+/******************************************************************************/
+/******************************************************************************/
+
+.Lstrmm_kernel_L2_BEGIN:   // less than 2 left in N direction
+
+	mov	counterJ , origN
+	tst	counterJ , #3
+	ble	.Lstrmm_kernel_L999
+
+	tst	counterJ , #2
+	ble	.Lstrmm_kernel_L1_BEGIN
+
+	mov	pCRow0, pC			// pCRow0 = pC
+
+	add	pC,pC,LDC, lsl #1
+
+#if defined(LEFT)
+	mov	tempOffset, offset
+#endif
+	mov	pA, origPA			// pA = A
+
+/******************************************************************************/
+
+.Lstrmm_kernel_L2_M8_BEGIN:
+
+	mov	counterI, origM
+	asr 	counterI, counterI, #3		// counterI = counterI / 8
+	cmp	counterI,#0
+	ble	.Lstrmm_kernel_L2_M4_BEGIN
+
+.Lstrmm_kernel_L2_M8_20:
+
+	INIT8x2
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	mov	pB, origPB
+#else
+	mov	pB, origPB
+	lsl	temp, tempOffset, #5
+	add	pA, pA, temp
+	lsl	temp, tempOffset, #3
+	add	pB, pB, temp
+#endif
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#elif defined(LEFT)
+	add	tempK, tempOffset, #8
+#else
+	add	tempK, tempOffset, #2
+#endif
+	asr	counterL , tempK, #3		// counterL = counterL / 8
+	cmp	counterL,#0
+	ble	.Lstrmm_kernel_L2_M8_40
+	.align 5
+
+.Lstrmm_kernel_L2_M8_22:
+	KERNEL8x2_SUB
+	KERNEL8x2_SUB
+	KERNEL8x2_SUB
+	KERNEL8x2_SUB
+
+	KERNEL8x2_SUB
+	KERNEL8x2_SUB
+	KERNEL8x2_SUB
+	KERNEL8x2_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L2_M8_22
+
+
+.Lstrmm_kernel_L2_M8_40:
+
+	ands	counterL , tempK, #7		// counterL = counterL % 8
+	ble	.Lstrmm_kernel_L2_M8_100
+
+.Lstrmm_kernel_L2_M8_42:
+
+	KERNEL8x2_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L2_M8_42
+
+.Lstrmm_kernel_L2_M8_100:
+
+	SAVE8x2
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#if defined(LEFT)
+	sub	tempK, tempK, #8
+#else
+	sub	tempK, tempK, #2
+#endif
+	lsl	temp, tempK, #5
+	add	pA, pA, temp
+	lsl	temp, tempK, #3
+	add	pB, pB, temp
+#endif
+#if defined(LEFT)
+	add	tempOffset, tempOffset, #8
+#endif
+.Lstrmm_kernel_L2_M8_END:
+
+	subs	counterI, counterI, #1
+	bgt	.Lstrmm_kernel_L2_M8_20
+
+/******************************************************************************/
+
+.Lstrmm_kernel_L2_M4_BEGIN:
+
+	mov	counterI, origM
+	tst	counterI , #7
+	ble	.Lstrmm_kernel_L2_END
+
+	tst	counterI, #4
+	ble	.Lstrmm_kernel_L2_M2_BEGIN
+
+.Lstrmm_kernel_L2_M4_20:
+
+	INIT4x2
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	mov	pB, origPB
+#else
+	mov	pB, origPB
+	lsl	temp, tempOffset, #3
+	add	pB, pB, temp
+	lsl	temp, tempOffset, #4
+	add	pA, pA, temp
+#endif
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#elif defined(LEFT)
+	add	tempK, tempOffset, #4
+#else
+	add	tempK, tempOffset, #2
+#endif
+	asr	counterL , tempK, #3		// counterL = counterL / 8
+	cmp	counterL,#0
+	ble	.Lstrmm_kernel_L2_M4_40
+	.align 5
+
+.Lstrmm_kernel_L2_M4_22:
+	KERNEL4x2_SUB
+	KERNEL4x2_SUB
+	KERNEL4x2_SUB
+	KERNEL4x2_SUB
+
+	KERNEL4x2_SUB
+	KERNEL4x2_SUB
+	KERNEL4x2_SUB
+	KERNEL4x2_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L2_M4_22
+
+
+.Lstrmm_kernel_L2_M4_40:
+
+	ands	counterL , tempK, #7		// counterL = counterL % 8
+	ble	.Lstrmm_kernel_L2_M4_100
+
+.Lstrmm_kernel_L2_M4_42:
+
+	KERNEL4x2_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L2_M4_42
+
+.Lstrmm_kernel_L2_M4_100:
+
+	SAVE4x2
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#if defined(LEFT)
+	sub	tempK, tempK, #4
+#else
+	sub	tempK, tempK, #2
+#endif
+	lsl	temp, tempK, #4
+	add	pA, pA, temp
+	lsl	temp, tempK, #3
+	add	pB, pB, temp
+#endif
+#if defined(LEFT)
+	add	tempOffset, tempOffset, #4
+#endif
+.Lstrmm_kernel_L2_M4_END:
+
+/******************************************************************************/
+
+.Lstrmm_kernel_L2_M2_BEGIN:
+
+	mov	counterI, origM
+	tst	counterI , #3
+	ble	.Lstrmm_kernel_L2_END
+
+	tst	counterI, #2			// counterI = counterI / 2
+	ble	.Lstrmm_kernel_L2_M1_BEGIN
+
+.Lstrmm_kernel_L2_M2_20:
+
+	INIT2x2
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	mov	pB, origPB
+#else
+	mov	pB, origPB
+	lsl	temp, tempOffset, #3
+	add	pB, pB, temp
+	lsl	temp, tempOffset, #3
+	add	pA, pA, temp
+#endif
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#elif defined(LEFT)
+	add	tempK, tempOffset, #2
+#else
+	add	tempK, tempOffset, #2
+#endif
+	asr	counterL , tempK, #3		// counterL = counterL / 8
+        cmp	counterL,#0
+	ble	.Lstrmm_kernel_L2_M2_40
+
+.Lstrmm_kernel_L2_M2_22:
+
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+	KERNEL2x2_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L2_M2_22
+
+
+.Lstrmm_kernel_L2_M2_40:
+
+	ands	counterL , tempK, #7		// counterL = counterL % 8
+	ble	.Lstrmm_kernel_L2_M2_100
+
+.Lstrmm_kernel_L2_M2_42:
+
+	KERNEL2x2_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L2_M2_42
+
+.Lstrmm_kernel_L2_M2_100:
+
+	SAVE2x2
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#if defined(LEFT)
+	sub	tempK, tempK, #2
+#else
+	sub	tempK, tempK, #2
+#endif
+	lsl	temp, tempK, #3
+	add	pA, pA, temp
+	lsl	temp, tempK, #3
+	add	pB, pB, temp
+#endif
+#if defined(LEFT)
+	add	tempOffset, tempOffset, #2
+#endif
+
+.Lstrmm_kernel_L2_M2_END:
+
+/******************************************************************************/
+
+.Lstrmm_kernel_L2_M1_BEGIN:
+
+	tst	counterI, #1			// counterI = counterI % 2
+	ble	.Lstrmm_kernel_L2_END
+
+.Lstrmm_kernel_L2_M1_20:
+
+	INIT1x2
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	mov	pB, origPB
+#else
+	mov	pB, origPB
+	lsl	temp, tempOffset, #3
+	add	pB, pB, temp
+	lsl	temp, tempOffset, #2
+	add	pA, pA, temp
+#endif
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#elif defined(LEFT)
+	add	tempK, tempOffset, #1
+#else
+	add	tempK, tempOffset, #2
+#endif
+	asr 	counterL , tempK, #3		// counterL = counterL / 8
+        cmp     counterL, #0
+	ble	.Lstrmm_kernel_L2_M1_40
+
+.Lstrmm_kernel_L2_M1_22:
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+	KERNEL1x2_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L2_M1_22
+
+
+.Lstrmm_kernel_L2_M1_40:
+
+	ands	counterL , tempK, #7		// counterL = counterL % 8
+	ble	.Lstrmm_kernel_L2_M1_100
+
+.Lstrmm_kernel_L2_M1_42:
+
+	KERNEL1x2_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L2_M1_42
+
+.Lstrmm_kernel_L2_M1_100:
+
+	SAVE1x2
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#if defined(LEFT)
+	sub	tempK, tempK, #1
+#else
+	sub	tempK, tempK, #2
+#endif
+	lsl	temp, tempK, #2
+	add	pA, pA, temp
+	lsl	temp, tempK, #3
+	add	pB, pB, temp
+#endif
+#if defined(LEFT)
+	add	tempOffset, tempOffset, #1
+#endif
+.Lstrmm_kernel_L2_END:
+#if !defined(LEFT)
+	add	tempOffset, tempOffset, #2
+#endif
+	add	origPB, origPB, origK, lsl #3	// B = B + K * 2 * 4
+
+/******************************************************************************/
+/******************************************************************************/
+
+.Lstrmm_kernel_L1_BEGIN:
+
+	mov	counterJ , origN
+	tst	counterJ , #1
+	ble	.Lstrmm_kernel_L999 // done
+
+
+	mov	pCRow0, pC			// pCRow0 = C
+	add	pC , pC , LDC			// Update pC to point to next
+
+#if defined(LEFT)
+	mov	tempOffset, offset
+#endif
+	mov	pA, origPA			// pA = A
+
+/******************************************************************************/
+
+.Lstrmm_kernel_L1_M8_BEGIN:
+
+	mov	counterI, origM
+	asr 	counterI, counterI, #3
+	cmp	counterI, #0
+	ble	.Lstrmm_kernel_L1_M4_BEGIN
+
+.Lstrmm_kernel_L1_M8_20:
+
+	INIT8x1
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	mov	pB, origPB
+#else
+	mov	pB, origPB
+	lsl	temp, tempOffset, #5
+	add	pA, pA, temp
+	lsl	temp, tempOffset, #2
+	add	pB, pB, temp
+#endif
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#elif defined(LEFT)
+	add	tempK, tempOffset, #8
+#else
+	add	tempK, tempOffset, #1
+#endif
+	asr	counterL , tempK, #3		// counterL = counterL / 8
+	cmp	counterL , #0
+	ble	.Lstrmm_kernel_L1_M8_40
+	.align 5
+
+.Lstrmm_kernel_L1_M8_22:
+	KERNEL8x1_SUB
+	KERNEL8x1_SUB
+	KERNEL8x1_SUB
+	KERNEL8x1_SUB
+
+	KERNEL8x1_SUB
+	KERNEL8x1_SUB
+	KERNEL8x1_SUB
+	KERNEL8x1_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L1_M8_22
+
+
+.Lstrmm_kernel_L1_M8_40:
+
+	ands	counterL , tempK, #7		// counterL = counterL % 8
+	ble	.Lstrmm_kernel_L1_M8_100
+
+.Lstrmm_kernel_L1_M8_42:
+
+	KERNEL8x1_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L1_M8_42
+
+.Lstrmm_kernel_L1_M8_100:
+
+	SAVE8x1
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#if defined(LEFT)
+	sub	tempK, tempK, #8
+#else
+	sub	tempK, tempK, #1
+#endif
+	lsl	temp, tempK, #5
+	add	pA, pA, temp
+	lsl	temp, tempK, #2
+	add	pB, pB, temp
+#endif
+#if defined(LEFT)
+	add	tempOffset, tempOffset, #8
+#endif
+.Lstrmm_kernel_L1_M8_END:
+
+	subs	counterI, counterI, #1
+	bgt	.Lstrmm_kernel_L1_M8_20
+
+/******************************************************************************/
+
+.Lstrmm_kernel_L1_M4_BEGIN:
+
+	mov	counterI, origM
+	tst	counterI , #7
+	ble	.Lstrmm_kernel_L1_END
+
+	tst	counterI, #4
+	ble	.Lstrmm_kernel_L1_M2_BEGIN
+
+.Lstrmm_kernel_L1_M4_20:
+
+	INIT4x1
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	mov	pB, origPB
+#else
+	mov	pB, origPB
+	lsl	temp, tempOffset, #2
+	add	pB, pB, temp
+	lsl	temp, tempOffset, #4
+	add	pA, pA, temp
+#endif
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#elif defined(LEFT)
+	add	tempK, tempOffset, #4
+#else
+	add	tempK, tempOffset, #1
+#endif
+	asr	counterL , tempK, #3		// counterL = counterL / 8
+	cmp	counterL , #0
+	ble	.Lstrmm_kernel_L1_M4_40
+	.align 5
+
+.Lstrmm_kernel_L1_M4_22:
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+	KERNEL4x1_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L1_M4_22
+
+
+.Lstrmm_kernel_L1_M4_40:
+
+	ands	counterL , tempK, #7		// counterL = counterL % 8
+	ble	.Lstrmm_kernel_L1_M4_100
+
+.Lstrmm_kernel_L1_M4_42:
+
+	KERNEL4x1_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L1_M4_42
+
+.Lstrmm_kernel_L1_M4_100:
+
+	SAVE4x1
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#if defined(LEFT)
+	sub	tempK, tempK, #4
+#else
+	sub	tempK, tempK, #1
+#endif
+	lsl	temp, tempK, #4
+	add	pA, pA, temp
+	lsl	temp, tempK, #2
+	add	pB, pB, temp
+#endif
+#if defined(LEFT)
+	add	tempOffset, tempOffset, #4
+#endif
+.Lstrmm_kernel_L1_M4_END:
+
+/******************************************************************************/
+
+.Lstrmm_kernel_L1_M2_BEGIN:
+
+	mov	counterI, origM
+	tst	counterI , #3
+	ble	.Lstrmm_kernel_L1_END
+
+	tst	counterI, #2			// counterI = counterI / 2
+	ble	.Lstrmm_kernel_L1_M1_BEGIN
+
+.Lstrmm_kernel_L1_M2_20:
+
+	INIT2x1
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	mov	pB, origPB
+#else
+	mov	pB, origPB
+	lsl	temp, tempOffset, #2
+	add	pB, pB, temp
+	lsl	temp, tempOffset, #3
+	add	pA, pA, temp
+#endif
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#elif defined(LEFT)
+	add	tempK, tempOffset, #2
+#else
+	add	tempK, tempOffset, #1
+#endif
+	asr 	counterL , tempK, #3		// counterL = counterL / 8
+	cmp	counterL , #0
+	ble	.Lstrmm_kernel_L1_M2_40
+
+.Lstrmm_kernel_L1_M2_22:
+
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+	KERNEL2x1_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L1_M2_22
+
+
+.Lstrmm_kernel_L1_M2_40:
+
+	ands	counterL , tempK, #7		// counterL = counterL % 8
+	ble	.Lstrmm_kernel_L1_M2_100
+
+.Lstrmm_kernel_L1_M2_42:
+
+	KERNEL2x1_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L1_M2_42
+
+.Lstrmm_kernel_L1_M2_100:
+
+	SAVE2x1
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#if defined(LEFT)
+	sub	tempK, tempK, #2
+#else
+	sub	tempK, tempK, #1
+#endif
+	lsl	temp, tempK, #3
+	add	pA, pA, temp
+	lsl	temp, tempK, #2
+	add	pB, pB, temp
+#endif
+#if defined(LEFT)
+	add	tempOffset, tempOffset, #2
+#endif
+.Lstrmm_kernel_L1_M2_END:
+
+/******************************************************************************/
+
+.Lstrmm_kernel_L1_M1_BEGIN:
+
+	tst	counterI, #1			// counterI = counterI % 2
+	ble	.Lstrmm_kernel_L1_END
+
+.Lstrmm_kernel_L1_M1_20:
+
+	INIT1x1
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+	mov	pB, origPB
+#else
+	mov	pB, origPB
+	lsl	temp, tempOffset, #2
+	add	pB, pB, temp
+	lsl	temp, tempOffset, #2
+	add	pA, pA, temp
+#endif
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+	sub	tempK, origK, tempOffset
+#elif defined(LEFT)
+	add	tempK, tempOffset, #1
+#else
+	add	tempK, tempOffset, #1
+#endif
+	asr 	counterL , tempK, #3		// counterL = counterL / 8
+	cmp	counterL , #0
+	ble	.Lstrmm_kernel_L1_M1_40
+
+.Lstrmm_kernel_L1_M1_22:
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+	KERNEL1x1_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L1_M1_22
+
+
+.Lstrmm_kernel_L1_M1_40:
+
+	ands	counterL , tempK, #7		// counterL = counterL % 8
+	ble	.Lstrmm_kernel_L1_M1_100
+
+.Lstrmm_kernel_L1_M1_42:
+
+	KERNEL1x1_SUB
+
+	subs	counterL, counterL, #1
+	bgt	.Lstrmm_kernel_L1_M1_42
+
+.Lstrmm_kernel_L1_M1_100:
+
+	SAVE1x1
+
+.Lstrmm_kernel_L1_END:
+
+/******************************************************************************/
+
+.Lstrmm_kernel_L999:
+	mov	x0, #0				// set return value
+	ldp	d8, d9, [sp, #(0 * 16)]
+	ldp	d10, d11, [sp, #(1 * 16)]
+	ldp	d12, d13, [sp, #(2 * 16)]
+	ldp	d14, d15, [sp, #(3 * 16)]
+	ldp	d16, d17, [sp, #(4 * 16)]
+	ldp	x18, x19, [sp, #(5 * 16)]
+	ldp	x20, x21, [sp, #(6 * 16)]
+	ldp	x22, x23, [sp, #(7 * 16)]
+	ldp	x24, x25, [sp, #(8 * 16)]
+	ldp	x26, x27, [sp, #(9 * 16)]
+	ldr	x28, [sp, #(10 * 16)]
+	add	sp, sp, #(11*16)
+	ret
+
+	EPILOGUE
+
diff --git a/kernel/arm64/zamax.S b/kernel/arm64/zamax.S
index c2c0a5374..3b49c4fff 100644
--- a/kernel/arm64/zamax.S
+++ b/kernel/arm64/zamax.S
@@ -137,10 +137,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	fadd	v3.2d, v3.2d, v4.2d
 #if defined(USE_MIN)
 	fmin	v1.2d, v1.2d, v3.2d
-	fminp	MAXF, v1.2d
+	fminp	TMPF, v1.2d
 #else
 	fmax	v1.2d, v1.2d, v3.2d
-	fmaxp	MAXF, v1.2d
+	fmaxp	TMPF, v1.2d
 #endif
 #endif
 	fcmp	MAXF, TMPF
diff --git a/kernel/generic/gemm_beta.c b/kernel/generic/gemm_beta.c
index fa9d7680d..ccb772cc7 100644
--- a/kernel/generic/gemm_beta.c
+++ b/kernel/generic/gemm_beta.c
@@ -39,7 +39,7 @@
 #include "common.h"
 
 int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT beta,
-	  FLOAT *dummy2, BLASLONG dummy3, FLOAT *dummy4, BLASLONG dummy5,
+	  IFLOAT *dummy2, BLASLONG dummy3, IFLOAT *dummy4, BLASLONG dummy5,
 	  FLOAT *c, BLASLONG ldc){
 
 
diff --git a/kernel/generic/gemm_ncopy_2.c b/kernel/generic/gemm_ncopy_2.c
index b728c713f..415860f81 100644
--- a/kernel/generic/gemm_ncopy_2.c
+++ b/kernel/generic/gemm_ncopy_2.c
@@ -39,10 +39,10 @@
 #include <stdio.h>
 #include "common.h"
 
-int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, FLOAT *b){
+int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b){
   BLASLONG i, j;
-  FLOAT *a_offset, *a_offset1, *a_offset2;
-  FLOAT *b_offset;
+  IFLOAT *a_offset, *a_offset1, *a_offset2;
+  IFLOAT *b_offset;
 
   a_offset = a;
   b_offset = b;
diff --git a/kernel/generic/gemm_tcopy_2.c b/kernel/generic/gemm_tcopy_2.c
index 5695b13c2..b4aa4de57 100644
--- a/kernel/generic/gemm_tcopy_2.c
+++ b/kernel/generic/gemm_tcopy_2.c
@@ -39,11 +39,11 @@
 #include <stdio.h>
 #include "common.h"
 
-int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, FLOAT *b){
+int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b){
   BLASLONG i, j;
 
-  FLOAT *a_offset, *a_offset1, *a_offset2;
-  FLOAT *b_offset, *b_offset1, *b_offset2;
+  IFLOAT *a_offset, *a_offset1, *a_offset2;
+  IFLOAT *b_offset, *b_offset1, *b_offset2;
 
   a_offset = a;
   b_offset = b;
diff --git a/kernel/generic/gemmkernel_2x2.c b/kernel/generic/gemmkernel_2x2.c
index 01f1c67b5..cc7bb8e48 100644
--- a/kernel/generic/gemmkernel_2x2.c
+++ b/kernel/generic/gemmkernel_2x2.c
@@ -1,13 +1,32 @@
 #include "common.h"
-int CNAME(BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alpha,FLOAT* ba,FLOAT* bb,FLOAT* C,BLASLONG ldc
+#if defined(HALF) && defined(HALFCONVERSION)
+static float
+bfloat16tof32 (bfloat16 f16)
+{
+  float result = 0;
+  unsigned short* q = (unsigned short*)(&result);
+#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+  q[0] = f16;
+#else
+  q[1] = f16;
+#endif
+  return result;
+}
+#define BF16TOF32(x) (bfloat16tof32(x))
+#else
+#define BF16TOF32(x) x
+#endif
+int CNAME(BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alpha,IFLOAT* ba,IFLOAT* bb,FLOAT* C,BLASLONG ldc
 #ifdef TRMMKERNEL
 		,BLASLONG offset
 #endif
 		)
 {
    BLASLONG i,j,k;
-   FLOAT *C0,*C1,*ptrba,*ptrbb;
-   FLOAT res0,res1,res2,res3,load0,load1,load2,load3,load4,load5,load6,load7;
+   FLOAT *C0,*C1;
+   IFLOAT *ptrba,*ptrbb;
+   FLOAT res0,res1,res2,res3;
+   IFLOAT load0,load1,load2,load3,load4,load5,load6,load7;
    for (j=0; j<bn/2; j+=1)
      {
         C0 = C;
@@ -24,36 +43,36 @@ int CNAME(BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alpha,FLOAT* ba,FLOAT* bb,FL
                {
                   load0 = ptrba[2*0+0];
                   load1 = ptrbb[2*0+0];
-                  res0 = res0+load0*load1;
+                  res0 = res0+BF16TOF32(load0)*BF16TOF32(load1);
                   load2 = ptrba[2*0+1];
-                  res1 = res1+load2*load1;
+                  res1 = res1+BF16TOF32(load2)*BF16TOF32(load1);
                   load3 = ptrbb[2*0+1];
-                  res2 = res2+load0*load3;
-                  res3 = res3+load2*load3;
+                  res2 = res2+BF16TOF32(load0)*BF16TOF32(load3);
+                  res3 = res3+BF16TOF32(load2)*BF16TOF32(load3);
                   load4 = ptrba[2*1+0];
                   load5 = ptrbb[2*1+0];
-                  res0 = res0+load4*load5;
+                  res0 = res0+BF16TOF32(load4)*BF16TOF32(load5);
                   load6 = ptrba[2*1+1];
-                  res1 = res1+load6*load5;
+                  res1 = res1+BF16TOF32(load6)*BF16TOF32(load5);
                   load7 = ptrbb[2*1+1];
-                  res2 = res2+load4*load7;
-                  res3 = res3+load6*load7;
+                  res2 = res2+BF16TOF32(load4)*BF16TOF32(load7);
+                  res3 = res3+BF16TOF32(load6)*BF16TOF32(load7);
                   load0 = ptrba[2*2+0];
                   load1 = ptrbb[2*2+0];
-                  res0 = res0+load0*load1;
+                  res0 = res0+BF16TOF32(load0)*BF16TOF32(load1);
                   load2 = ptrba[2*2+1];
-                  res1 = res1+load2*load1;
+                  res1 = res1+BF16TOF32(load2)*BF16TOF32(load1);
                   load3 = ptrbb[2*2+1];
-                  res2 = res2+load0*load3;
-                  res3 = res3+load2*load3;
+                  res2 = res2+BF16TOF32(load0)*BF16TOF32(load3);
+                  res3 = res3+BF16TOF32(load2)*BF16TOF32(load3);
                   load4 = ptrba[2*3+0];
                   load5 = ptrbb[2*3+0];
-                  res0 = res0+load4*load5;
+                  res0 = res0+BF16TOF32(load4)*BF16TOF32(load5);
                   load6 = ptrba[2*3+1];
-                  res1 = res1+load6*load5;
+                  res1 = res1+BF16TOF32(load6)*BF16TOF32(load5);
                   load7 = ptrbb[2*3+1];
-                  res2 = res2+load4*load7;
-                  res3 = res3+load6*load7;
+                  res2 = res2+BF16TOF32(load4)*BF16TOF32(load7);
+                  res3 = res3+BF16TOF32(load6)*BF16TOF32(load7);
                   ptrba = ptrba+8;
                   ptrbb = ptrbb+8;
                }
@@ -61,12 +80,12 @@ int CNAME(BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alpha,FLOAT* ba,FLOAT* bb,FL
                {
                   load0 = ptrba[2*0+0];
                   load1 = ptrbb[2*0+0];
-                  res0 = res0+load0*load1;
+                  res0 = res0+BF16TOF32(load0)*BF16TOF32(load1);
                   load2 = ptrba[2*0+1];
-                  res1 = res1+load2*load1;
+                  res1 = res1+BF16TOF32(load2)*BF16TOF32(load1);
                   load3 = ptrbb[2*0+1];
-                  res2 = res2+load0*load3;
-                  res3 = res3+load2*load3;
+                  res2 = res2+BF16TOF32(load0)*BF16TOF32(load3);
+                  res3 = res3+BF16TOF32(load2)*BF16TOF32(load3);
                   ptrba = ptrba+2;
                   ptrbb = ptrbb+2;
                }
@@ -90,9 +109,9 @@ int CNAME(BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alpha,FLOAT* ba,FLOAT* bb,FL
                {
                   load0 = ptrba[0+0];
                   load1 = ptrbb[2*0+0];
-                  res0 = res0+load0*load1;
+                  res0 = res0+BF16TOF32(load0)*BF16TOF32(load1);
                   load2 = ptrbb[2*0+1];
-                  res1 = res1+load0*load2;
+                  res1 = res1+BF16TOF32(load0)*BF16TOF32(load2);
                   ptrba = ptrba+1;
                   ptrbb = ptrbb+2;
                }
@@ -121,9 +140,9 @@ int CNAME(BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alpha,FLOAT* ba,FLOAT* bb,FL
                {
                   load0 = ptrba[2*0+0];
                   load1 = ptrbb[0+0];
-                  res0 = res0+load0*load1;
+                  res0 = res0+BF16TOF32(load0)*BF16TOF32(load1);
                   load2 = ptrba[2*0+1];
-                  res1 = res1+load2*load1;
+                  res1 = res1+BF16TOF32(load2)*BF16TOF32(load1);
                   ptrba = ptrba+2;
                   ptrbb = ptrbb+1;
                }
@@ -141,7 +160,7 @@ int CNAME(BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alpha,FLOAT* ba,FLOAT* bb,FL
                {
                   load0 = ptrba[0+0];
                   load1 = ptrbb[0+0];
-                  res0 = res0+load0*load1;
+                  res0 = res0+BF16TOF32(load0)*BF16TOF32(load1);
                   ptrba = ptrba+1;
                   ptrbb = ptrbb+1;
                }
diff --git a/kernel/mips/KERNEL.1004K b/kernel/mips/KERNEL.MIPS1004K
similarity index 100%
rename from kernel/mips/KERNEL.1004K
rename to kernel/mips/KERNEL.MIPS1004K
diff --git a/kernel/mips/KERNEL.MIPS24K b/kernel/mips/KERNEL.MIPS24K
new file mode 100644
index 000000000..67135356e
--- /dev/null
+++ b/kernel/mips/KERNEL.MIPS24K
@@ -0,0 +1 @@
+include $(KERNELDIR)/KERNEL.P5600
diff --git a/kernel/mips64/KERNEL.generic b/kernel/mips64/KERNEL.generic
new file mode 100644
index 000000000..17f2ef976
--- /dev/null
+++ b/kernel/mips64/KERNEL.generic
@@ -0,0 +1,160 @@
+SGEMM_BETA = ../generic/gemm_beta.c
+DGEMM_BETA = ../generic/gemm_beta.c
+CGEMM_BETA = ../generic/zgemm_beta.c
+ZGEMM_BETA = ../generic/zgemm_beta.c
+
+STRMMKERNEL	= ../generic/trmmkernel_2x2.c
+DTRMMKERNEL	= ../generic/trmmkernel_2x2.c
+CTRMMKERNEL	= ../generic/ztrmmkernel_2x2.c
+ZTRMMKERNEL	= ../generic/ztrmmkernel_2x2.c
+
+SGEMMKERNEL    =  ../generic/gemmkernel_2x2.c
+SGEMMONCOPY    =  ../generic/gemm_ncopy_2.c
+SGEMMOTCOPY    =  ../generic/gemm_tcopy_2.c
+SGEMMONCOPYOBJ =  sgemm_oncopy.o
+SGEMMOTCOPYOBJ =  sgemm_otcopy.o
+
+DGEMMKERNEL    =  ../generic/gemmkernel_2x2.c
+DGEMMONCOPY    = ../generic/gemm_ncopy_2.c
+DGEMMOTCOPY    = ../generic/gemm_tcopy_2.c
+DGEMMONCOPYOBJ = dgemm_oncopy.o
+DGEMMOTCOPYOBJ = dgemm_otcopy.o
+
+CGEMMKERNEL    = ../generic/zgemmkernel_2x2.c
+CGEMMONCOPY    = ../generic/zgemm_ncopy_2.c
+CGEMMOTCOPY    = ../generic/zgemm_tcopy_2.c
+CGEMMONCOPYOBJ =  cgemm_oncopy.o
+CGEMMOTCOPYOBJ =  cgemm_otcopy.o
+
+ZGEMMKERNEL    = ../generic/zgemmkernel_2x2.c
+ZGEMMONCOPY    = ../generic/zgemm_ncopy_2.c
+ZGEMMOTCOPY    = ../generic/zgemm_tcopy_2.c
+ZGEMMONCOPYOBJ =  zgemm_oncopy.o
+ZGEMMOTCOPYOBJ =  zgemm_otcopy.o
+
+STRSMKERNEL_LN	=  ../generic/trsm_kernel_LN.c
+STRSMKERNEL_LT	=  ../generic/trsm_kernel_LT.c
+STRSMKERNEL_RN	=  ../generic/trsm_kernel_RN.c
+STRSMKERNEL_RT	=  ../generic/trsm_kernel_RT.c
+
+DTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
+DTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
+DTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
+DTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
+
+CTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
+CTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
+CTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
+CTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
+
+ZTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
+ZTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
+ZTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
+ZTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
+
+#Pure C for other kernels
+SAMAXKERNEL  = ../mips/amax.c
+DAMAXKERNEL  = ../mips/amax.c
+CAMAXKERNEL  = ../mips/zamax.c
+ZAMAXKERNEL  = ../mips/zamax.c
+
+SAMINKERNEL  = ../mips/amin.c
+DAMINKERNEL  = ../mips/amin.c
+CAMINKERNEL  = ../mips/zamin.c
+ZAMINKERNEL  = ../mips/zamin.c
+
+SMAXKERNEL   = ../mips/max.c
+DMAXKERNEL   = ../mips/max.c
+
+SMINKERNEL   = ../mips/min.c
+DMINKERNEL   = ../mips/min.c
+
+ISAMAXKERNEL = ../mips/iamax.c
+IDAMAXKERNEL = ../mips/iamax.c
+ICAMAXKERNEL = ../mips/izamax.c
+IZAMAXKERNEL = ../mips/izamax.c
+
+ISAMINKERNEL = ../mips/iamin.c
+IDAMINKERNEL = ../mips/iamin.c
+ICAMINKERNEL = ../mips/izamin.c
+IZAMINKERNEL = ../mips/izamin.c
+
+ISMAXKERNEL  = ../mips/imax.c
+IDMAXKERNEL  = ../mips/imax.c
+
+ISMINKERNEL  = ../mips/imin.c
+IDMINKERNEL  = ../mips/imin.c
+
+SASUMKERNEL  = ../mips/asum.c
+DASUMKERNEL  = ../mips/asum.c
+CASUMKERNEL  = ../mips/zasum.c
+ZASUMKERNEL  = ../mips/zasum.c
+
+SSUMKERNEL  = ../mips/sum.c
+DSUMKERNEL  = ../mips/sum.c
+CSUMKERNEL  = ../mips/zsum.c
+ZSUMKERNEL  = ../mips/zsum.c
+
+SAXPYKERNEL  = ../mips/axpy.c
+DAXPYKERNEL  = ../mips/axpy.c
+CAXPYKERNEL  = ../mips/zaxpy.c
+ZAXPYKERNEL  = ../mips/zaxpy.c
+
+SCOPYKERNEL  = ../mips/copy.c
+DCOPYKERNEL  = ../mips/copy.c
+CCOPYKERNEL  = ../mips/zcopy.c
+ZCOPYKERNEL  = ../mips/zcopy.c
+
+SDOTKERNEL   = ../mips/dot.c
+DDOTKERNEL   = ../mips/dot.c
+CDOTKERNEL   = ../mips/zdot.c
+ZDOTKERNEL   = ../mips/zdot.c
+
+SNRM2KERNEL  = ../mips/nrm2.c
+DNRM2KERNEL  = ../mips/nrm2.c
+CNRM2KERNEL  = ../mips/znrm2.c
+ZNRM2KERNEL  = ../mips/znrm2.c
+
+SROTKERNEL   = ../mips/rot.c
+DROTKERNEL   = ../mips/rot.c
+CROTKERNEL   = ../mips/zrot.c
+ZROTKERNEL   = ../mips/zrot.c
+
+SSCALKERNEL  = ../mips/scal.c
+DSCALKERNEL  = ../mips/scal.c
+CSCALKERNEL  = ../mips/zscal.c
+ZSCALKERNEL  = ../mips/zscal.c
+
+SSWAPKERNEL  = ../mips/swap.c
+DSWAPKERNEL  = ../mips/swap.c
+CSWAPKERNEL  = ../mips/zswap.c
+ZSWAPKERNEL  = ../mips/zswap.c
+
+SGEMVNKERNEL = ../mips/gemv_n.c
+DGEMVNKERNEL = ../mips/gemv_n.c
+CGEMVNKERNEL = ../mips/zgemv_n.c
+ZGEMVNKERNEL = ../mips/zgemv_n.c
+
+SGEMVTKERNEL = ../mips/gemv_t.c
+DGEMVTKERNEL = ../mips/gemv_t.c
+CGEMVTKERNEL = ../mips/zgemv_t.c
+ZGEMVTKERNEL = ../mips/zgemv_t.c
+
+SSYMV_U_KERNEL =  ../generic/symv_k.c
+SSYMV_L_KERNEL =  ../generic/symv_k.c
+DSYMV_U_KERNEL =  ../generic/symv_k.c
+DSYMV_L_KERNEL =  ../generic/symv_k.c
+QSYMV_U_KERNEL =  ../generic/symv_k.c
+QSYMV_L_KERNEL =  ../generic/symv_k.c
+CSYMV_U_KERNEL =  ../generic/zsymv_k.c
+CSYMV_L_KERNEL =  ../generic/zsymv_k.c
+ZSYMV_U_KERNEL =  ../generic/zsymv_k.c
+ZSYMV_L_KERNEL =  ../generic/zsymv_k.c
+XSYMV_U_KERNEL =  ../generic/zsymv_k.c
+XSYMV_L_KERNEL =  ../generic/zsymv_k.c
+
+ZHEMV_U_KERNEL =  ../generic/zhemv_k.c
+ZHEMV_L_KERNEL =  ../generic/zhemv_k.c
+
+CGEMM3MKERNEL    = ../generic/zgemm3mkernel_dump.c
+ZGEMM3MKERNEL    = ../generic/zgemm3mkernel_dump.c
diff --git a/kernel/power/KERNEL.POWER8 b/kernel/power/KERNEL.POWER8
index c7867012b..7fba5b4d6 100644
--- a/kernel/power/KERNEL.POWER8
+++ b/kernel/power/KERNEL.POWER8
@@ -12,7 +12,7 @@ SGEMMKERNEL    =  sgemm_kernel_16x8_power8.S
 SGEMMINCOPY    = ../generic/gemm_ncopy_16.c
 SGEMMITCOPY    = sgemm_tcopy_16_power8.S
 SGEMMONCOPY    =  ../generic/gemm_ncopy_8.c
-SGEMMOTCOPY    = sgemm_tcopy_8_power8.S 
+SGEMMOTCOPY    = sgemm_tcopy_8_power8.S
 SGEMMINCOPYOBJ =  sgemm_incopy$(TSUFFIX).$(SUFFIX)
 SGEMMITCOPYOBJ =  sgemm_itcopy$(TSUFFIX).$(SUFFIX)
 SGEMMONCOPYOBJ =  sgemm_oncopy$(TSUFFIX).$(SUFFIX)
@@ -232,3 +232,11 @@ QCABS_KERNEL	= ../generic/cabs.c
 #Dump kernel
 CGEMM3MKERNEL    = ../generic/zgemm3mkernel_dump.c
 ZGEMM3MKERNEL    = ../generic/zgemm3mkernel_dump.c
+
+ifeq ($(__BYTE_ORDER__)$(ELF_VERSION),__ORDER_BIG_ENDIAN__2)
+IDAMAXKERNEL  = ../arm/iamax.c
+IDAMINKERNEL  = ../arm/iamin.c
+IZAMAXKERNEL  = ../arm/izamax.c
+IZAMINKERNEL  = ../arm/izamin.c
+endif
+
diff --git a/kernel/power/KERNEL.POWER9 b/kernel/power/KERNEL.POWER9
index aabb5d976..ab8fbfcd9 100644
--- a/kernel/power/KERNEL.POWER9
+++ b/kernel/power/KERNEL.POWER9
@@ -16,7 +16,7 @@ SGEMMKERNEL    =  sgemm_kernel_power9.S
 SGEMMINCOPY    = ../generic/gemm_ncopy_16.c
 SGEMMITCOPY    = sgemm_tcopy_16_power8.S
 SGEMMONCOPY    =  ../generic/gemm_ncopy_8.c
-SGEMMOTCOPY    = sgemm_tcopy_8_power8.S 
+SGEMMOTCOPY    = sgemm_tcopy_8_power8.S
 SGEMMINCOPYOBJ =  sgemm_incopy$(TSUFFIX).$(SUFFIX)
 SGEMMITCOPYOBJ =  sgemm_itcopy$(TSUFFIX).$(SUFFIX)
 SGEMMONCOPYOBJ =  sgemm_oncopy$(TSUFFIX).$(SUFFIX)
diff --git a/kernel/power/KERNEL.PPCG4 b/kernel/power/KERNEL.PPCG4
index f615754bb..54660b54d 100644
--- a/kernel/power/KERNEL.PPCG4
+++ b/kernel/power/KERNEL.PPCG4
@@ -20,8 +20,10 @@ ZAXPYKERNEL	= zaxpy_ppc440.S
 
 SDOTKERNEL	=  dot_ppc440.S
 DDOTKERNEL	=  dot_ppc440.S
-CDOTKERNEL	= zdot_ppc440.S
-ZDOTKERNEL	= zdot_ppc440.S
+#CDOTKERNEL	= zdot_ppc440.S
+#ZDOTKERNEL	= zdot_ppc440.S
+CDOTKERNEL	= ../arm/zdot.c
+ZDOTKERNEL	= ../arm/zdot.c
 
 ISAMAXKERNEL	=  iamax_ppc440.S
 IDAMAXKERNEL	=  iamax_ppc440.S
@@ -52,8 +54,11 @@ ZNRM2KERNEL	= znrm2_ppc440.S
 
 SROTKERNEL	=  rot_ppc440.S
 DROTKERNEL	=  rot_ppc440.S
-CROTKERNEL	= zrot_ppc440.S
-ZROTKERNEL	= zrot_ppc440.S
+#CROTKERNEL	= zrot_ppc440.S
+#ZROTKERNEL	= zrot_ppc440.S
+CROTKERNEL	= ../arm/zrot.c
+ZROTKERNEL	= ../arm/zrot.c
+
 
 SSCALKERNEL	=  scal_ppc440.S
 DSCALKERNEL	=  scal_ppc440.S
@@ -78,13 +83,18 @@ DGEMMINCOPYOBJ =
 DGEMMITCOPYOBJ =
 DGEMMONCOPYOBJ =  dgemm_oncopy$(TSUFFIX).$(SUFFIX)
 DGEMMOTCOPYOBJ =  dgemm_otcopy$(TSUFFIX).$(SUFFIX)
-CGEMMKERNEL    =  zgemm_kernel_altivec_g4.S
-CGEMMINCOPY    =  ../generic/zgemm_ncopy_8.c
-CGEMMITCOPY    =  ../generic/zgemm_tcopy_8.c
+#CGEMMKERNEL    =  zgemm_kernel_altivec_g4.S
+#CGEMMINCOPY    =  ../generic/zgemm_ncopy_8.c
+#CGEMMITCOPY    =  ../generic/zgemm_tcopy_8.c
+CGEMMKERNEL    =  zgemm_kernel.S
+CGEMMINCOPY    =
+CGEMMONCOPY    =
 CGEMMONCOPY    =  ../generic/zgemm_ncopy_2.c
 CGEMMOTCOPY    =  ../generic/zgemm_tcopy_2.c
-CGEMMINCOPYOBJ =  cgemm_incopy$(TSUFFIX).$(SUFFIX)
-CGEMMITCOPYOBJ =  cgemm_itcopy$(TSUFFIX).$(SUFFIX)
+CGEMMINCOPYOBJ =  
+#cgemm_incopy$(TSUFFIX).$(SUFFIX)
+CGEMMITCOPYOBJ =  
+#cgemm_itcopy$(TSUFFIX).$(SUFFIX)
 CGEMMONCOPYOBJ =  cgemm_oncopy$(TSUFFIX).$(SUFFIX)
 CGEMMOTCOPYOBJ =  cgemm_otcopy$(TSUFFIX).$(SUFFIX)
 ZGEMMKERNEL    =  zgemm_kernel_g4.S
diff --git a/kernel/power/caxpy_power9.S b/kernel/power/caxpy_power9.S
index 844cacd50..b4733ff9f 100644
--- a/kernel/power/caxpy_power9.S
+++ b/kernel/power/caxpy_power9.S
@@ -13,7 +13,11 @@
 
 	PROLOGUE
 
+#ifdef CONJ
+caxpyc_k:
+#else
 caxpy_k:
+#endif
 .LCF0:
 0:	addis 2,12,.TOC.-.LCF0@ha
 	addi 2,2,.TOC.-.LCF0@l
diff --git a/kernel/power/icamax_power9.S b/kernel/power/icamax_power9.S
index 2968b3f8b..bf6ab6e82 100644
--- a/kernel/power/icamax_power9.S
+++ b/kernel/power/icamax_power9.S
@@ -1,3 +1,4 @@
+/*
 	.file	"icamax.c"
 	.abiversion 2
 	.section	".text"
@@ -5,6 +6,12 @@
 	.p2align 4,,15
 	.globl icamax_k
 	.type	icamax_k, @function
+*/
+#define ASSEMBLER
+#include "common.h"
+
+        PROLOGUE
+
 icamax_k:
 .LCF0:
 0:	addis 2,12,.TOC.-.LCF0@ha
diff --git a/kernel/power/icamin_power9.S b/kernel/power/icamin_power9.S
index 8eaa79f33..58a3c53a3 100644
--- a/kernel/power/icamin_power9.S
+++ b/kernel/power/icamin_power9.S
@@ -1,3 +1,4 @@
+/*
 	.file	"icamin.c"
 	.abiversion 2
 	.section	".text"
@@ -5,6 +6,12 @@
 	.p2align 4,,15
 	.globl icamin_k
 	.type	icamin_k, @function
+*/
+#define ASSEMBLER
+#include "common.h"
+
+        PROLOGUE
+
 icamin_k:
 .LCF0:
 0:	addis 2,12,.TOC.-.LCF0@ha
diff --git a/kernel/power/isamax_power9.S b/kernel/power/isamax_power9.S
index 9df1e773c..259c996fc 100644
--- a/kernel/power/isamax_power9.S
+++ b/kernel/power/isamax_power9.S
@@ -1,3 +1,4 @@
+/*
 	.file	"isamax.c"
 	.abiversion 2
 	.section	".text"
@@ -5,6 +6,12 @@
 	.p2align 4,,15
 	.globl isamax_k
 	.type	isamax_k, @function
+*/
+#define ASSEMBLER
+#include "common.h"
+
+        PROLOGUE
+
 isamax_k:
 .LCF0:
 0:	addis 2,12,.TOC.-.LCF0@ha
diff --git a/kernel/power/isamin_power9.S b/kernel/power/isamin_power9.S
index 0475edf46..36486ff02 100644
--- a/kernel/power/isamin_power9.S
+++ b/kernel/power/isamin_power9.S
@@ -1,3 +1,4 @@
+/*
 	.file	"isamin.c"
 	.abiversion 2
 	.section	".text"
@@ -5,6 +6,12 @@
 	.p2align 4,,15
 	.globl isamin_k
 	.type	isamin_k, @function
+*/
+#define ASSEMBLER
+#include "common.h"
+
+        PROLOGUE
+
 isamin_k:
 .LCF0:
 0:	addis 2,12,.TOC.-.LCF0@ha
diff --git a/kernel/setparam-ref.c b/kernel/setparam-ref.c
index 3c71c778e..d3aa030c1 100644
--- a/kernel/setparam-ref.c
+++ b/kernel/setparam-ref.c
@@ -53,6 +53,65 @@ gotoblas_t TABLE_NAME = {
 
   GEMM_DEFAULT_OFFSET_A, GEMM_DEFAULT_OFFSET_B, GEMM_DEFAULT_ALIGN,
 
+#ifdef BUILD_HALF
+  0, 0, 0,
+  SHGEMM_DEFAULT_UNROLL_M, SHGEMM_DEFAULT_UNROLL_N,
+#ifdef SHGEMM_DEFAULT_UNROLL_MN
+ SHGEMM_DEFAULT_UNROLL_MN,
+#else
+ MAX(SHGEMM_DEFAULT_UNROLL_M, SHGEMM_DEFAULT_UNROLL_N),
+#endif
+
+  samax_kTS,  samin_kTS,  smax_kTS,  smin_kTS,
+  isamax_kTS, isamin_kTS, ismax_kTS, ismin_kTS,
+  snrm2_kTS,  sasum_kTS, ssum_kTS, scopy_kTS, sdot_kTS,
+  dsdot_kTS,
+  srot_kTS,   saxpy_kTS,  sscal_kTS, sswap_kTS,
+  sgemv_nTS,  sgemv_tTS, sger_kTS,
+  ssymv_LTS, ssymv_UTS,
+
+  shgemm_kernelTS, shgemm_betaTS,
+#if SHGEMM_DEFAULT_UNROLL_M != SHGEMM_DEFAULT_UNROLL_N
+  shgemm_incopyTS, shgemm_itcopyTS,
+#else
+  shgemm_oncopyTS, shgemm_otcopyTS,
+#endif
+  shgemm_oncopyTS, shgemm_otcopyTS,
+
+  strsm_kernel_LNTS, strsm_kernel_LTTS, strsm_kernel_RNTS, strsm_kernel_RTTS,
+#if SGEMM_DEFAULT_UNROLL_M != SGEMM_DEFAULT_UNROLL_N
+  strsm_iunucopyTS, strsm_iunncopyTS, strsm_iutucopyTS, strsm_iutncopyTS,
+  strsm_ilnucopyTS, strsm_ilnncopyTS, strsm_iltucopyTS, strsm_iltncopyTS,
+#else
+  strsm_ounucopyTS, strsm_ounncopyTS, strsm_outucopyTS, strsm_outncopyTS,
+  strsm_olnucopyTS, strsm_olnncopyTS, strsm_oltucopyTS, strsm_oltncopyTS,
+#endif
+  strsm_ounucopyTS, strsm_ounncopyTS, strsm_outucopyTS, strsm_outncopyTS,
+  strsm_olnucopyTS, strsm_olnncopyTS, strsm_oltucopyTS, strsm_oltncopyTS,
+  strmm_kernel_RNTS, strmm_kernel_RTTS, strmm_kernel_LNTS, strmm_kernel_LTTS,
+#if SGEMM_DEFAULT_UNROLL_M != SGEMM_DEFAULT_UNROLL_N
+  strmm_iunucopyTS, strmm_iunncopyTS, strmm_iutucopyTS, strmm_iutncopyTS,
+  strmm_ilnucopyTS, strmm_ilnncopyTS, strmm_iltucopyTS, strmm_iltncopyTS,
+#else
+  strmm_ounucopyTS, strmm_ounncopyTS, strmm_outucopyTS, strmm_outncopyTS,
+  strmm_olnucopyTS, strmm_olnncopyTS, strmm_oltucopyTS, strmm_oltncopyTS,
+#endif
+  strmm_ounucopyTS, strmm_ounncopyTS, strmm_outucopyTS, strmm_outncopyTS,
+  strmm_olnucopyTS, strmm_olnncopyTS, strmm_oltucopyTS, strmm_oltncopyTS,
+#if SGEMM_DEFAULT_UNROLL_M != SGEMM_DEFAULT_UNROLL_N
+  ssymm_iutcopyTS, ssymm_iltcopyTS,
+#else
+  ssymm_outcopyTS, ssymm_oltcopyTS,
+#endif
+  ssymm_outcopyTS, ssymm_oltcopyTS,
+
+#ifndef NO_LAPACK
+  sneg_tcopyTS, slaswp_ncopyTS,
+#else
+  NULL,NULL,
+#endif
+#endif
+
   0, 0, 0,
   SGEMM_DEFAULT_UNROLL_M, SGEMM_DEFAULT_UNROLL_N,
 #ifdef SGEMM_DEFAULT_UNROLL_MN
@@ -648,16 +707,25 @@ gotoblas_t TABLE_NAME = {
 
 #if defined(ARCH_ARM64)
 static void init_parameter(void) {
+#if defined(BUILD_HALF)
+  TABLE_NAME.shgemm_p = SHGEMM_DEFAULT_P;
+#endif
   TABLE_NAME.sgemm_p = SGEMM_DEFAULT_P;
   TABLE_NAME.dgemm_p = DGEMM_DEFAULT_P;
   TABLE_NAME.cgemm_p = CGEMM_DEFAULT_P;
   TABLE_NAME.zgemm_p = ZGEMM_DEFAULT_P;
 
+#if defined(BUILD_HALF)
+  TABLE_NAME.shgemm_q = SHGEMM_DEFAULT_Q;
+#endif
   TABLE_NAME.sgemm_q = SGEMM_DEFAULT_Q;
   TABLE_NAME.dgemm_q = DGEMM_DEFAULT_Q;
   TABLE_NAME.cgemm_q = CGEMM_DEFAULT_Q;
   TABLE_NAME.zgemm_q = ZGEMM_DEFAULT_Q;
 
+#if defined(BUILD_HALF)
+  TABLE_NAME.shgemm_r = SHGEMM_DEFAULT_R;
+#endif
   TABLE_NAME.sgemm_r = SGEMM_DEFAULT_R;
   TABLE_NAME.dgemm_r = DGEMM_DEFAULT_R;
   TABLE_NAME.cgemm_r = CGEMM_DEFAULT_R;
@@ -721,17 +789,26 @@ static void init_parameter(void) {
 #if defined(ARCH_POWER)
 static void init_parameter(void) {
 
+#ifdef BUILD_HALF
+  TABLE_NAME.shgemm_p = SHGEMM_DEFAULT_P;
+#endif
   TABLE_NAME.sgemm_p = SGEMM_DEFAULT_P;
   TABLE_NAME.dgemm_p = DGEMM_DEFAULT_P;
   TABLE_NAME.cgemm_p = CGEMM_DEFAULT_P;
   TABLE_NAME.zgemm_p = ZGEMM_DEFAULT_P;
 
+#ifdef BUILD_HALF
+  TABLE_NAME.shgemm_r = SHGEMM_DEFAULT_R;
+#endif
   TABLE_NAME.sgemm_r = SGEMM_DEFAULT_R;
   TABLE_NAME.dgemm_r = DGEMM_DEFAULT_R;
   TABLE_NAME.cgemm_r = CGEMM_DEFAULT_R;
   TABLE_NAME.zgemm_r = ZGEMM_DEFAULT_R;
 
 
+#ifdef BUILD_HALF
+  TABLE_NAME.shgemm_q = SHGEMM_DEFAULT_Q;
+#endif
   TABLE_NAME.sgemm_q = SGEMM_DEFAULT_Q;
   TABLE_NAME.dgemm_q = DGEMM_DEFAULT_Q;
   TABLE_NAME.cgemm_q = CGEMM_DEFAULT_Q;
@@ -741,17 +818,26 @@ static void init_parameter(void) {
 
 #if defined(ARCH_ZARCH)
 static void init_parameter(void) {
+#ifdef BUILD_HALF
+	TABLE_NAME.shgemm_p = SHGEMM_DEFAULT_P;
+#endif
 	TABLE_NAME.sgemm_p = SGEMM_DEFAULT_P;
 	TABLE_NAME.dgemm_p = DGEMM_DEFAULT_P;
 	TABLE_NAME.cgemm_p = CGEMM_DEFAULT_P;
 	TABLE_NAME.zgemm_p = ZGEMM_DEFAULT_P;
 
+#ifdef BUILD_HALF
+	TABLE_NAME.shgemm_r = SHGEMM_DEFAULT_R;
+#endif
 	TABLE_NAME.sgemm_r = SGEMM_DEFAULT_R;
 	TABLE_NAME.dgemm_r = DGEMM_DEFAULT_R;
 	TABLE_NAME.cgemm_r = CGEMM_DEFAULT_R;
 	TABLE_NAME.zgemm_r = ZGEMM_DEFAULT_R;
 
 
+#ifdef BUILD_HALF
+	TABLE_NAME.shgemm_q = SHGEMM_DEFAULT_Q;
+#endif
 	TABLE_NAME.sgemm_q = SGEMM_DEFAULT_Q;
 	TABLE_NAME.dgemm_q = DGEMM_DEFAULT_Q;
 	TABLE_NAME.cgemm_q = CGEMM_DEFAULT_Q;
@@ -891,6 +977,11 @@ static void init_parameter(void) {
   (void) l2; /* dirty trick to suppress unused variable warning for targets */
              /* where the GEMM unrolling parameters do not depend on l2 */
   
+#ifdef BUILD_HALF
+  TABLE_NAME.shgemm_p = SHGEMM_DEFAULT_P;
+  TABLE_NAME.shgemm_r = SHGEMM_DEFAULT_R;
+  TABLE_NAME.shgemm_q = SHGEMM_DEFAULT_Q;
+#endif
   TABLE_NAME.sgemm_q = SGEMM_DEFAULT_Q;
   TABLE_NAME.dgemm_q = DGEMM_DEFAULT_Q;
   TABLE_NAME.cgemm_q = CGEMM_DEFAULT_Q;
diff --git a/kernel/x86_64/KERNEL.HASWELL b/kernel/x86_64/KERNEL.HASWELL
index f6ca5c2d5..ef8b36a57 100644
--- a/kernel/x86_64/KERNEL.HASWELL
+++ b/kernel/x86_64/KERNEL.HASWELL
@@ -77,10 +77,10 @@ ZGEMMITCOPYOBJ =  zgemm_itcopy$(TSUFFIX).$(SUFFIX)
 ZGEMMONCOPYOBJ =  zgemm_oncopy$(TSUFFIX).$(SUFFIX)
 ZGEMMOTCOPYOBJ =  zgemm_otcopy$(TSUFFIX).$(SUFFIX)
 
-STRSMKERNEL_LN	=  ../generic/trsm_kernel_LN.c
-STRSMKERNEL_LT	=  ../generic/trsm_kernel_LT.c
-STRSMKERNEL_RN	=  ../generic/trsm_kernel_RN.c
-STRSMKERNEL_RT	=  ../generic/trsm_kernel_RT.c
+STRSMKERNEL_LN	=  strsm_kernel_8x4_haswell_LN.c
+STRSMKERNEL_LT	=  strsm_kernel_8x4_haswell_LT.c
+STRSMKERNEL_RN	=  strsm_kernel_8x4_haswell_RN.c
+STRSMKERNEL_RT	=  strsm_kernel_8x4_haswell_RT.c
 
 DTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
 DTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
diff --git a/kernel/x86_64/KERNEL.SKYLAKEX b/kernel/x86_64/KERNEL.SKYLAKEX
index 9b3c83e42..448aee074 100644
--- a/kernel/x86_64/KERNEL.SKYLAKEX
+++ b/kernel/x86_64/KERNEL.SKYLAKEX
@@ -1,11 +1,15 @@
 include $(KERNELDIR)/KERNEL.HASWELL
 
-SGEMMKERNEL    =  sgemm_kernel_16x4_skylakex_2.c
+SGEMMKERNEL    =  sgemm_kernel_16x4_skylakex_3.c
 STRMMKERNEL    =  sgemm_kernel_16x4_skylakex_2.c
 SGEMMINCOPY    =  ../generic/gemm_ncopy_16.c
 SGEMMITCOPY    =  sgemm_tcopy_16_skylakex.c
 SGEMMONCOPY    =  sgemm_ncopy_4_skylakex.c
 SGEMMOTCOPY    =  ../generic/gemm_tcopy_4.c
+STRSMKERNEL_LN =  ../generic/trsm_kernel_LN.c
+STRSMKERNEL_LT =  ../generic/trsm_kernel_LT.c
+STRSMKERNEL_RN =  ../generic/trsm_kernel_RN.c
+STRSMKERNEL_RT =  ../generic/trsm_kernel_RT.c
 
 DGEMMKERNEL    =  dgemm_kernel_16x2_skylakex.c
 DTRMMKERNEL    =  dgemm_kernel_16x2_skylakex.c
@@ -20,3 +24,6 @@ DGEMM_BETA = dgemm_beta_skylakex.c
 
 CGEMMKERNEL    =  cgemm_kernel_8x2_skylakex.c
 ZGEMMKERNEL    =  zgemm_kernel_4x2_skylakex.c
+
+CSCALKERNEL    = ../arm/zscal.c
+ZSCALKERNEL    = ../arm/zscal.c
diff --git a/kernel/x86_64/KERNEL.ZEN b/kernel/x86_64/KERNEL.ZEN
index 1cd02db74..7bb308fea 100644
--- a/kernel/x86_64/KERNEL.ZEN
+++ b/kernel/x86_64/KERNEL.ZEN
@@ -74,10 +74,10 @@ ZGEMMITCOPYOBJ =  zgemm_itcopy$(TSUFFIX).$(SUFFIX)
 ZGEMMONCOPYOBJ =  zgemm_oncopy$(TSUFFIX).$(SUFFIX)
 ZGEMMOTCOPYOBJ =  zgemm_otcopy$(TSUFFIX).$(SUFFIX)
 
-STRSMKERNEL_LN	=  ../generic/trsm_kernel_LN.c
-STRSMKERNEL_LT	=  ../generic/trsm_kernel_LT.c
-STRSMKERNEL_RN	=  ../generic/trsm_kernel_RN.c
-STRSMKERNEL_RT	=  ../generic/trsm_kernel_RT.c
+STRSMKERNEL_LN	=  strsm_kernel_8x4_haswell_LN.c
+STRSMKERNEL_LT	=  strsm_kernel_8x4_haswell_LT.c
+STRSMKERNEL_RN	=  strsm_kernel_8x4_haswell_RN.c
+STRSMKERNEL_RT	=  strsm_kernel_8x4_haswell_RT.c
 
 DTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
 DTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
diff --git a/kernel/x86_64/copy_sse2.S b/kernel/x86_64/copy_sse2.S
index 200daafd9..a5ab2ea91 100644
--- a/kernel/x86_64/copy_sse2.S
+++ b/kernel/x86_64/copy_sse2.S
@@ -54,7 +54,7 @@
 #ifdef OPTERON
 #define LOAD(OFFSET, ADDR, REG)		xorps	REG, REG; addpd	OFFSET(ADDR), REG
 #else
-#define LOAD(OFFSET, ADDR, REG)		movaps	OFFSET(ADDR), REG
+#define LOAD(OFFSET, ADDR, REG)		movups	OFFSET(ADDR), REG
 #endif
 
 	PROLOGUE
@@ -104,14 +104,14 @@
 	sarq	$4, %rax
 	jle	.L13
 
-	movaps	-16 * SIZE(X), %xmm0
-	movaps	-14 * SIZE(X), %xmm1
-	movaps	-12 * SIZE(X), %xmm2
-	movaps	-10 * SIZE(X), %xmm3
-	movaps	 -8 * SIZE(X), %xmm4
-	movaps	 -6 * SIZE(X), %xmm5
-	movaps	 -4 * SIZE(X), %xmm6
-	movaps	 -2 * SIZE(X), %xmm7
+	movups	-16 * SIZE(X), %xmm0
+	movups	-14 * SIZE(X), %xmm1
+	movups	-12 * SIZE(X), %xmm2
+	movups	-10 * SIZE(X), %xmm3
+	movups	 -8 * SIZE(X), %xmm4
+	movups	 -6 * SIZE(X), %xmm5
+	movups	 -4 * SIZE(X), %xmm6
+	movups	 -2 * SIZE(X), %xmm7
 
 	decq	%rax
 	jle .L12
@@ -122,36 +122,36 @@
 	PREFETCHW (PREFETCHSIZE +  0) - PREOFFSET(Y)
 #endif
 
-	movaps	%xmm0, -16 * SIZE(Y)
+	movups	%xmm0, -16 * SIZE(Y)
 	LOAD( 0 * SIZE, X, %xmm0)
-	movaps	%xmm1, -14 * SIZE(Y)
+	movups	%xmm1, -14 * SIZE(Y)
 	LOAD( 2 * SIZE, X, %xmm1)
 
 #ifdef PREFETCH
 	PREFETCH (PREFETCHSIZE +  0) - PREOFFSET(X)
 #endif
 
-	movaps	%xmm2, -12 * SIZE(Y)
+	movups	%xmm2, -12 * SIZE(Y)
 	LOAD( 4 * SIZE, X, %xmm2)
-	movaps	%xmm3, -10 * SIZE(Y)
+	movups	%xmm3, -10 * SIZE(Y)
 	LOAD( 6 * SIZE, X, %xmm3)
 
 #if defined(PREFETCHW) && !defined(FETCH128)
 	PREFETCHW (PREFETCHSIZE +  64) - PREOFFSET(Y)
 #endif
 
-	movaps	%xmm4, -8 * SIZE(Y)
+	movups	%xmm4, -8 * SIZE(Y)
 	LOAD( 8 * SIZE, X, %xmm4)
-	movaps	%xmm5, -6 * SIZE(Y)
+	movups	%xmm5, -6 * SIZE(Y)
 	LOAD(10 * SIZE, X, %xmm5)
 
 #if defined(PREFETCH) && !defined(FETCH128)
 	PREFETCH (PREFETCHSIZE +  64) - PREOFFSET(X)
 #endif
 
-	movaps	%xmm6, -4 * SIZE(Y)
+	movups	%xmm6, -4 * SIZE(Y)
 	LOAD(12 * SIZE, X, %xmm6)
-	movaps	%xmm7, -2 * SIZE(Y)
+	movups	%xmm7, -2 * SIZE(Y)
 	LOAD(14 * SIZE, X, %xmm7)
 
 	subq	$-16 * SIZE, Y
@@ -161,14 +161,14 @@
 	ALIGN_3
 
 .L12:
-	movaps	%xmm0, -16 * SIZE(Y)
-	movaps	%xmm1, -14 * SIZE(Y)
-	movaps	%xmm2, -12 * SIZE(Y)
-	movaps	%xmm3, -10 * SIZE(Y)
-	movaps	%xmm4,  -8 * SIZE(Y)
-	movaps	%xmm5,  -6 * SIZE(Y)
-	movaps	%xmm6,  -4 * SIZE(Y)
-	movaps	%xmm7,  -2 * SIZE(Y)
+	movups	%xmm0, -16 * SIZE(Y)
+	movups	%xmm1, -14 * SIZE(Y)
+	movups	%xmm2, -12 * SIZE(Y)
+	movups	%xmm3, -10 * SIZE(Y)
+	movups	%xmm4,  -8 * SIZE(Y)
+	movups	%xmm5,  -6 * SIZE(Y)
+	movups	%xmm6,  -4 * SIZE(Y)
+	movups	%xmm7,  -2 * SIZE(Y)
 
 	subq	$-16 * SIZE, Y
 	subq	$-16 * SIZE, X
@@ -179,15 +179,15 @@
 	jle	.L14
 	ALIGN_3
 
-	movaps	-16 * SIZE(X), %xmm0
-	movaps	-14 * SIZE(X), %xmm1
-	movaps	-12 * SIZE(X), %xmm2
-	movaps	-10 * SIZE(X), %xmm3
+	movups	-16 * SIZE(X), %xmm0
+	movups	-14 * SIZE(X), %xmm1
+	movups	-12 * SIZE(X), %xmm2
+	movups	-10 * SIZE(X), %xmm3
 
-	movaps	%xmm0, -16 * SIZE(Y)
-	movaps	%xmm1, -14 * SIZE(Y)
-	movaps	%xmm2, -12 * SIZE(Y)
-	movaps	%xmm3, -10 * SIZE(Y)
+	movups	%xmm0, -16 * SIZE(Y)
+	movups	%xmm1, -14 * SIZE(Y)
+	movups	%xmm2, -12 * SIZE(Y)
+	movups	%xmm3, -10 * SIZE(Y)
 
 	addq	$8 * SIZE, X
 	addq	$8 * SIZE, Y
@@ -198,11 +198,11 @@
 	jle	.L15
 	ALIGN_3
 
-	movaps	-16 * SIZE(X), %xmm0
-	movaps	-14 * SIZE(X), %xmm1
+	movups	-16 * SIZE(X), %xmm0
+	movups	-14 * SIZE(X), %xmm1
 
-	movaps	%xmm0, -16 * SIZE(Y)
-	movaps	%xmm1, -14 * SIZE(Y)
+	movups	%xmm0, -16 * SIZE(Y)
+	movups	%xmm1, -14 * SIZE(Y)
 
 	addq	$4 * SIZE, X
 	addq	$4 * SIZE, Y
@@ -213,8 +213,8 @@
 	jle	.L16
 	ALIGN_3
 
-	movaps	-16 * SIZE(X), %xmm0
-	movaps	%xmm0, -16 * SIZE(Y)
+	movups	-16 * SIZE(X), %xmm0
+	movups	%xmm0, -16 * SIZE(Y)
 
 	addq	$2 * SIZE, X
 	addq	$2 * SIZE, Y
@@ -246,13 +246,13 @@
 	sarq	$4, %rax
 	jle	.L23
 
-	movaps	-15 * SIZE(X), %xmm1
-	movaps	-13 * SIZE(X), %xmm2
-	movaps	-11 * SIZE(X), %xmm3
-	movaps	 -9 * SIZE(X), %xmm4
-	movaps	 -7 * SIZE(X), %xmm5
-	movaps	 -5 * SIZE(X), %xmm6
-	movaps	 -3 * SIZE(X), %xmm7
+	movups	-15 * SIZE(X), %xmm1
+	movups	-13 * SIZE(X), %xmm2
+	movups	-11 * SIZE(X), %xmm3
+	movups	 -9 * SIZE(X), %xmm4
+	movups	 -7 * SIZE(X), %xmm5
+	movups	 -5 * SIZE(X), %xmm6
+	movups	 -3 * SIZE(X), %xmm7
 
 	decq	%rax
 	jle .L22
@@ -264,11 +264,11 @@
 #endif
 
 	SHUFPD_1 %xmm1, %xmm0
-	movaps	%xmm0, -16 * SIZE(Y)
+	movups	%xmm0, -16 * SIZE(Y)
 	LOAD(-1 * SIZE, X, %xmm0)
 
 	SHUFPD_1 %xmm2, %xmm1
-	movaps	%xmm1, -14 * SIZE(Y)
+	movups	%xmm1, -14 * SIZE(Y)
 	LOAD( 1 * SIZE, X, %xmm1)
 
 #ifdef PREFETCH
@@ -276,11 +276,11 @@
 #endif
 
 	SHUFPD_1 %xmm3, %xmm2
-	movaps	%xmm2, -12 * SIZE(Y)
+	movups	%xmm2, -12 * SIZE(Y)
 	LOAD( 3 * SIZE, X, %xmm2)
 
 	SHUFPD_1 %xmm4, %xmm3
-	movaps	%xmm3, -10 * SIZE(Y)
+	movups	%xmm3, -10 * SIZE(Y)
 	LOAD( 5 * SIZE, X, %xmm3)
 
 #if defined(PREFETCHW) && !defined(FETCH128)
@@ -288,11 +288,11 @@
 #endif
 
 	SHUFPD_1 %xmm5, %xmm4
-	movaps	%xmm4,  -8 * SIZE(Y)
+	movups	%xmm4,  -8 * SIZE(Y)
 	LOAD( 7 * SIZE, X, %xmm4)
 
 	SHUFPD_1 %xmm6, %xmm5
-	movaps	%xmm5, -6 * SIZE(Y)
+	movups	%xmm5, -6 * SIZE(Y)
 	LOAD( 9 * SIZE, X, %xmm5)
 
 #if defined(PREFETCH) && !defined(FETCH128)
@@ -300,11 +300,11 @@
 #endif
 
 	SHUFPD_1 %xmm7, %xmm6
-	movaps	%xmm6, -4 * SIZE(Y)
+	movups	%xmm6, -4 * SIZE(Y)
 	LOAD(11 * SIZE, X, %xmm6)
 
 	SHUFPD_1 %xmm0, %xmm7
-	movaps	%xmm7, -2 * SIZE(Y)
+	movups	%xmm7, -2 * SIZE(Y)
 	LOAD(13 * SIZE, X, %xmm7)
 
 	subq	$-16 * SIZE, X
@@ -315,26 +315,26 @@
 
 .L22:
 	SHUFPD_1 %xmm1, %xmm0
-	movaps	%xmm0, -16 * SIZE(Y)
+	movups	%xmm0, -16 * SIZE(Y)
 	LOAD(-1 * SIZE, X, %xmm0)
 
 	SHUFPD_1 %xmm2, %xmm1
-	movaps	%xmm1, -14 * SIZE(Y)
+	movups	%xmm1, -14 * SIZE(Y)
 
 	SHUFPD_1 %xmm3, %xmm2
-	movaps	%xmm2, -12 * SIZE(Y)
+	movups	%xmm2, -12 * SIZE(Y)
 	SHUFPD_1 %xmm4, %xmm3
-	movaps	%xmm3, -10 * SIZE(Y)
+	movups	%xmm3, -10 * SIZE(Y)
 
 	SHUFPD_1 %xmm5, %xmm4
-	movaps	%xmm4,  -8 * SIZE(Y)
+	movups	%xmm4,  -8 * SIZE(Y)
 	SHUFPD_1 %xmm6, %xmm5
-	movaps	%xmm5,  -6 * SIZE(Y)
+	movups	%xmm5,  -6 * SIZE(Y)
 
 	SHUFPD_1 %xmm7, %xmm6
-	movaps	%xmm6,  -4 * SIZE(Y)
+	movups	%xmm6,  -4 * SIZE(Y)
 	SHUFPD_1 %xmm0, %xmm7
-	movaps	%xmm7,  -2 * SIZE(Y)
+	movups	%xmm7,  -2 * SIZE(Y)
 
 	subq	$-16 * SIZE, X
 	subq	$-16 * SIZE, Y
@@ -345,24 +345,24 @@
 	jle	.L24
 	ALIGN_3
 
-	movaps	-15 * SIZE(X), %xmm1
-	movaps	-13 * SIZE(X), %xmm2
-	movaps	-11 * SIZE(X), %xmm3
-	movaps	 -9 * SIZE(X), %xmm8
+	movups	-15 * SIZE(X), %xmm1
+	movups	-13 * SIZE(X), %xmm2
+	movups	-11 * SIZE(X), %xmm3
+	movups	 -9 * SIZE(X), %xmm8
 
 	SHUFPD_1 %xmm1, %xmm0
-	movaps	%xmm0, -16 * SIZE(Y)
+	movups	%xmm0, -16 * SIZE(Y)
 
 	SHUFPD_1 %xmm2, %xmm1
-	movaps	%xmm1, -14 * SIZE(Y)
+	movups	%xmm1, -14 * SIZE(Y)
 
 	SHUFPD_1 %xmm3, %xmm2
-	movaps	%xmm2, -12 * SIZE(Y)
+	movups	%xmm2, -12 * SIZE(Y)
 
 	SHUFPD_1 %xmm8, %xmm3
-	movaps	%xmm3, -10 * SIZE(Y)
+	movups	%xmm3, -10 * SIZE(Y)
 
-	movaps	%xmm8, %xmm0
+	movups	%xmm8, %xmm0
 
 	addq	$8 * SIZE, X
 	addq	$8 * SIZE, Y
@@ -373,15 +373,15 @@
 	jle	.L25
 	ALIGN_3
 
-	movaps	-15 * SIZE(X), %xmm1
-	movaps	-13 * SIZE(X), %xmm2
+	movups	-15 * SIZE(X), %xmm1
+	movups	-13 * SIZE(X), %xmm2
 
 	SHUFPD_1 %xmm1, %xmm0
 	SHUFPD_1 %xmm2, %xmm1
 
-	movaps	%xmm0, -16 * SIZE(Y)
-	movaps	%xmm1, -14 * SIZE(Y)
-	movaps	%xmm2, %xmm0
+	movups	%xmm0, -16 * SIZE(Y)
+	movups	%xmm1, -14 * SIZE(Y)
+	movups	%xmm2, %xmm0
 
 	addq	$4 * SIZE, X
 	addq	$4 * SIZE, Y
@@ -392,10 +392,10 @@
 	jle	.L26
 	ALIGN_3
 
-	movaps	-15 * SIZE(X), %xmm1
+	movups	-15 * SIZE(X), %xmm1
 	SHUFPD_1 %xmm1, %xmm0
 
-	movaps	%xmm0, -16 * SIZE(Y)
+	movups	%xmm0, -16 * SIZE(Y)
 
 	addq	$2 * SIZE, X
 	addq	$2 * SIZE, Y
@@ -424,14 +424,14 @@
 	sarq	$4, %rax
 	jle	.L23
 
-	movaps	-16 * SIZE(X), %xmm0
-	movaps	-14 * SIZE(X), %xmm1
-	movaps	-12 * SIZE(X), %xmm2
-	movaps	-10 * SIZE(X), %xmm3
-	movaps	 -8 * SIZE(X), %xmm4
-	movaps	 -6 * SIZE(X), %xmm5
-	movaps	 -4 * SIZE(X), %xmm6
-	movaps	 -2 * SIZE(X), %xmm7
+	movups	-16 * SIZE(X), %xmm0
+	movups	-14 * SIZE(X), %xmm1
+	movups	-12 * SIZE(X), %xmm2
+	movups	-10 * SIZE(X), %xmm3
+	movups	 -8 * SIZE(X), %xmm4
+	movups	 -6 * SIZE(X), %xmm5
+	movups	 -4 * SIZE(X), %xmm6
+	movups	 -2 * SIZE(X), %xmm7
 
 	decq	%rax
 	jle .L22
@@ -515,16 +515,16 @@
 	jle	.L24
 	ALIGN_3
 
-	movaps	-16 * SIZE(X), %xmm0
+	movups	-16 * SIZE(X), %xmm0
 	movlps	%xmm0, -16 * SIZE(Y)
 	movhps	%xmm0, -15 * SIZE(Y)
-	movaps	-14 * SIZE(X), %xmm1
+	movups	-14 * SIZE(X), %xmm1
 	movlps	%xmm1, -14 * SIZE(Y)
 	movhps	%xmm1, -13 * SIZE(Y)
-	movaps	-12 * SIZE(X), %xmm2
+	movups	-12 * SIZE(X), %xmm2
 	movlps	%xmm2, -12 * SIZE(Y)
 	movhps	%xmm2, -11 * SIZE(Y)
-	movaps	-10 * SIZE(X), %xmm3
+	movups	-10 * SIZE(X), %xmm3
 	movlps	%xmm3, -10 * SIZE(Y)
 	movhps	%xmm3,  -9 * SIZE(Y)
 
@@ -537,10 +537,10 @@
 	jle	.L25
 	ALIGN_3
 
-	movaps	-16 * SIZE(X), %xmm0
+	movups	-16 * SIZE(X), %xmm0
 	movlps	%xmm0, -16 * SIZE(Y)
 	movhps	%xmm0, -15 * SIZE(Y)
-	movaps	-14 * SIZE(X), %xmm1
+	movups	-14 * SIZE(X), %xmm1
 	movlps	%xmm1, -14 * SIZE(Y)
 	movhps	%xmm1, -13 * SIZE(Y)
 
@@ -553,7 +553,7 @@
 	jle	.L26
 	ALIGN_3
 
-	movaps	-16 * SIZE(X), %xmm0
+	movups	-16 * SIZE(X), %xmm0
 	movlps	%xmm0, -16 * SIZE(Y)
 	movhps	%xmm0, -15 * SIZE(Y)
 
diff --git a/kernel/x86_64/dgemm_kernel_16x2_skylakex.c b/kernel/x86_64/dgemm_kernel_16x2_skylakex.c
index 416ace59b..9f2bf24e2 100644
--- a/kernel/x86_64/dgemm_kernel_16x2_skylakex.c
+++ b/kernel/x86_64/dgemm_kernel_16x2_skylakex.c
@@ -54,40 +54,40 @@
     #define kernel_kstart_n10(mdim,updk) ""
     #define kernel_kstart_n12(mdim,updk) ""
     #define kernel_kend_n4(mdim) "xorq %3,%3;"\
-      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(0,8)\
-      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(16,24)
+      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(0)\
+      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(16)
     #define kernel_kend_n6(mdim) "xorq %3,%3;"\
-      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(0,8) acc_kend_nc3_k1m##mdim(0,8)\
-      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(16,24) acc_kend_nc3_k1m##mdim(16,24)\
-      loada_kend_k1m##mdim acc_kend_nc3_k1m##mdim(32,40)\
-      loada_kend_k1m##mdim acc_kend_nc3_k1m##mdim(48,56)
+      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(0) acc_kend_nc3_k1m##mdim(0)\
+      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(16) acc_kend_nc3_k1m##mdim(16)\
+      loada_kend_k1m##mdim acc_kend_nc3_k1m##mdim(32)\
+      loada_kend_k1m##mdim acc_kend_nc3_k1m##mdim(48)
     #define kernel_kend_n8(mdim) "xorq %3,%3;"\
-      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(0,8) acc_kend_nc3_k1m##mdim(0,8) acc_kend_nc4_k1m##mdim(0,8)\
-      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(16,24) acc_kend_nc3_k1m##mdim(16,24) acc_kend_nc4_k1m##mdim(16,24)\
-      loada_kend_k1m##mdim acc_kend_nc3_k1m##mdim(32,40) acc_kend_nc4_k1m##mdim(32,40)\
-      loada_kend_k1m##mdim acc_kend_nc3_k1m##mdim(48,56) acc_kend_nc4_k1m##mdim(48,56)\
-      loada_kend_k1m##mdim acc_kend_nc4_k1m##mdim(64,72)\
-      loada_kend_k1m##mdim acc_kend_nc4_k1m##mdim(80,88)
+      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(0) acc_kend_nc3_k1m##mdim(0) acc_kend_nc4_k1m##mdim(0)\
+      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(16) acc_kend_nc3_k1m##mdim(16) acc_kend_nc4_k1m##mdim(16)\
+      loada_kend_k1m##mdim acc_kend_nc3_k1m##mdim(32) acc_kend_nc4_k1m##mdim(32)\
+      loada_kend_k1m##mdim acc_kend_nc3_k1m##mdim(48) acc_kend_nc4_k1m##mdim(48)\
+      loada_kend_k1m##mdim acc_kend_nc4_k1m##mdim(64)\
+      loada_kend_k1m##mdim acc_kend_nc4_k1m##mdim(80)
     #define kernel_kend_n10(mdim) "xorq %3,%3;"\
-      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(0,8) acc_kend_nc3_k1m##mdim(0,8) acc_kend_nc4_k1m##mdim(0,8) acc_kend_nc5_k1m##mdim(0,8)\
-      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(16,24) acc_kend_nc3_k1m##mdim(16,24) acc_kend_nc4_k1m##mdim(16,24) acc_kend_nc5_k1m##mdim(16,24)\
-      loada_kend_k1m##mdim acc_kend_nc3_k1m##mdim(32,40) acc_kend_nc4_k1m##mdim(32,40) acc_kend_nc5_k1m##mdim(32,40)\
-      loada_kend_k1m##mdim acc_kend_nc3_k1m##mdim(48,56) acc_kend_nc4_k1m##mdim(48,56) acc_kend_nc5_k1m##mdim(48,56)\
-      loada_kend_k1m##mdim acc_kend_nc4_k1m##mdim(64,72) acc_kend_nc5_k1m##mdim(64,72)\
-      loada_kend_k1m##mdim acc_kend_nc4_k1m##mdim(80,88) acc_kend_nc5_k1m##mdim(80,88)\
-      loada_kend_k1m##mdim acc_kend_nc5_k1m##mdim(96,104)\
-      loada_kend_k1m##mdim acc_kend_nc5_k1m##mdim(112,120)
+      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(0) acc_kend_nc3_k1m##mdim(0) acc_kend_nc4_k1m##mdim(0) acc_kend_nc5_k1m##mdim(0)\
+      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(16) acc_kend_nc3_k1m##mdim(16) acc_kend_nc4_k1m##mdim(16) acc_kend_nc5_k1m##mdim(16)\
+      loada_kend_k1m##mdim acc_kend_nc3_k1m##mdim(32) acc_kend_nc4_k1m##mdim(32) acc_kend_nc5_k1m##mdim(32)\
+      loada_kend_k1m##mdim acc_kend_nc3_k1m##mdim(48) acc_kend_nc4_k1m##mdim(48) acc_kend_nc5_k1m##mdim(48)\
+      loada_kend_k1m##mdim acc_kend_nc4_k1m##mdim(64) acc_kend_nc5_k1m##mdim(64)\
+      loada_kend_k1m##mdim acc_kend_nc4_k1m##mdim(80) acc_kend_nc5_k1m##mdim(80)\
+      loada_kend_k1m##mdim acc_kend_nc5_k1m##mdim(96)\
+      loada_kend_k1m##mdim acc_kend_nc5_k1m##mdim(112)
     #define kernel_kend_n12(mdim) "xorq %3,%3;"\
-      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(0,8) acc_kend_nc3_k1m##mdim(0,8) acc_kend_nc4_k1m##mdim(0,8) acc_kend_nc5_k1m##mdim(0,8) acc_kend_nc6_k1m##mdim(0,8)\
-      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(16,24) acc_kend_nc3_k1m##mdim(16,24) acc_kend_nc4_k1m##mdim(16,24) acc_kend_nc5_k1m##mdim(16,24) acc_kend_nc6_k1m##mdim(16,24)\
-      loada_kend_k1m##mdim acc_kend_nc3_k1m##mdim(32,40) acc_kend_nc4_k1m##mdim(32,40) acc_kend_nc5_k1m##mdim(32,40) acc_kend_nc6_k1m##mdim(32,40)\
-      loada_kend_k1m##mdim acc_kend_nc3_k1m##mdim(48,56) acc_kend_nc4_k1m##mdim(48,56) acc_kend_nc5_k1m##mdim(48,56) acc_kend_nc6_k1m##mdim(48,56)\
-      loada_kend_k1m##mdim acc_kend_nc4_k1m##mdim(64,72) acc_kend_nc5_k1m##mdim(64,72) acc_kend_nc6_k1m##mdim(64,72)\
-      loada_kend_k1m##mdim acc_kend_nc4_k1m##mdim(80,88) acc_kend_nc5_k1m##mdim(80,88) acc_kend_nc6_k1m##mdim(80,88)\
-      loada_kend_k1m##mdim acc_kend_nc5_k1m##mdim(96,104) acc_kend_nc6_k1m##mdim(96,104)\
-      loada_kend_k1m##mdim acc_kend_nc5_k1m##mdim(112,120) acc_kend_nc6_k1m##mdim(112,120)\
-      loada_kend_k1m##mdim acc_kend_nc6_k1m##mdim(128,136)\
-      loada_kend_k1m##mdim acc_kend_nc6_k1m##mdim(144,152)
+      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(0) acc_kend_nc3_k1m##mdim(0) acc_kend_nc4_k1m##mdim(0) acc_kend_nc5_k1m##mdim(0) acc_kend_nc6_k1m##mdim(0)\
+      loada_kend_k1m##mdim acc_kend_nc2_k1m##mdim(16) acc_kend_nc3_k1m##mdim(16) acc_kend_nc4_k1m##mdim(16) acc_kend_nc5_k1m##mdim(16) acc_kend_nc6_k1m##mdim(16)\
+      loada_kend_k1m##mdim acc_kend_nc3_k1m##mdim(32) acc_kend_nc4_k1m##mdim(32) acc_kend_nc5_k1m##mdim(32) acc_kend_nc6_k1m##mdim(32)\
+      loada_kend_k1m##mdim acc_kend_nc3_k1m##mdim(48) acc_kend_nc4_k1m##mdim(48) acc_kend_nc5_k1m##mdim(48) acc_kend_nc6_k1m##mdim(48)\
+      loada_kend_k1m##mdim acc_kend_nc4_k1m##mdim(64) acc_kend_nc5_k1m##mdim(64) acc_kend_nc6_k1m##mdim(64)\
+      loada_kend_k1m##mdim acc_kend_nc4_k1m##mdim(80) acc_kend_nc5_k1m##mdim(80) acc_kend_nc6_k1m##mdim(80)\
+      loada_kend_k1m##mdim acc_kend_nc5_k1m##mdim(96) acc_kend_nc6_k1m##mdim(96)\
+      loada_kend_k1m##mdim acc_kend_nc5_k1m##mdim(112) acc_kend_nc6_k1m##mdim(112)\
+      loada_kend_k1m##mdim acc_kend_nc6_k1m##mdim(128)\
+      loada_kend_k1m##mdim acc_kend_nc6_k1m##mdim(144)
   #endif
 #else
   #define HEAD_SET_OFF(ndim) {}
@@ -129,18 +129,28 @@
   #define init_update_k(mdim) ""
   #define save_update_k(mdim) ""
 #endif
-    
+
 #define KERNEL_h_k1m16n1 \
   "vmovupd (%0),%%zmm1; vmovupd 64(%0),%%zmm2; addq $128,%0;"\
   "vbroadcastsd (%1),%%zmm3; vfmadd231pd %%zmm1,%%zmm3,%%zmm8; vfmadd231pd %%zmm2,%%zmm3,%%zmm9;"
 #define KERNEL_k1m16n1 KERNEL_h_k1m16n1 "addq $8,%1;"
-#define KERNEL_h_k1m16n2 KERNEL_h_k1m16n1\
+#ifdef BROADCAST_KERNEL
+ #define KERNEL_h_k1m16n2 KERNEL_h_k1m16n1\
   "vbroadcastsd 8(%1),%%zmm4; vfmadd231pd %%zmm1,%%zmm4,%%zmm10; vfmadd231pd %%zmm2,%%zmm4,%%zmm11;"
-#define KERNEL_k1m16n2 KERNEL_h_k1m16n2 "addq $16,%1;"
-#define unit_acc_gen_m16n2(c1_no,c2_no,c3_no,c4_no,boff1,boff2,...)\
+ #define unit_acc_gen_m16n2(c1_no,c2_no,c3_no,c4_no,boff1,...)\
   "vbroadcastsd "#boff1"("#__VA_ARGS__"),%%zmm3; vfmadd231pd %%zmm1,%%zmm3,%%zmm"#c1_no"; vfmadd231pd %%zmm2,%%zmm3,%%zmm"#c2_no";"\
-  "vbroadcastsd "#boff2"("#__VA_ARGS__"),%%zmm4; vfmadd231pd %%zmm1,%%zmm4,%%zmm"#c3_no"; vfmadd231pd %%zmm2,%%zmm4,%%zmm"#c4_no";"
-#define unit_acc_m16n2(c1_no,c2_no,c3_no,c4_no,...) unit_acc_gen_m16n2(c1_no,c2_no,c3_no,c4_no,0,8,__VA_ARGS__)
+  "vbroadcastsd "#boff1"+8("#__VA_ARGS__"),%%zmm4; vfmadd231pd %%zmm1,%%zmm4,%%zmm"#c3_no"; vfmadd231pd %%zmm2,%%zmm4,%%zmm"#c4_no";"
+ #define unit_acc_m16n2(c1_no,c2_no,c3_no,c4_no,...) unit_acc_gen_m16n2(c1_no,c2_no,c3_no,c4_no,0,__VA_ARGS__)
+#else
+ #define unit_acc_gen_m16n2(c1_no,c2_no,c3_no,c4_no,boff1,...)\
+  "vbroadcastf32x4 "#boff1"("#__VA_ARGS__"),%%zmm5; vfmadd231pd %%zmm1,%%zmm5,%%zmm"#c1_no"; vfmadd231pd %%zmm2,%%zmm5,%%zmm"#c2_no";"\
+  "vfmadd231pd %%zmm3,%%zmm5,%%zmm"#c3_no"; vfmadd231pd %%zmm4,%%zmm5,%%zmm"#c4_no";"
+ #define unit_acc_m16n2(c1_no,c2_no,c3_no,c4_no,...) unit_acc_gen_m16n2(c1_no,c2_no,c3_no,c4_no,0,__VA_ARGS__)
+ #define KERNEL_h_k1m16n2 \
+  "vmovddup (%0),%%zmm1; vmovddup 8(%0),%%zmm2; vmovddup 64(%0),%%zmm3; vmovddup 72(%0),%%zmm4; addq $128,%0;"\
+  unit_acc_m16n2(8,9,10,11,%1)
+#endif
+#define KERNEL_k1m16n2 KERNEL_h_k1m16n2 "addq $16,%1;"
 #define KERNEL_h_k1m16n4 KERNEL_h_k1m16n2 "prefetcht0 384(%0);" unit_acc_m16n2(12,13,14,15,%1,%%r12,1)
 #define KERNEL_k1m16n4 KERNEL_h_k1m16n4 "addq $16,%1;"
 #define KERNEL_k1m16n6 KERNEL_h_k1m16n4 unit_acc_m16n2(16,17,18,19,%1,%%r12,2) "addq $16,%1;"
@@ -151,24 +161,42 @@
 #define KERNEL_h_k1m16n12 KERNEL_h_k1m16n10 unit_acc_m16n2(28,29,30,31,%%r15,%%r12,2)
 #define KERNEL_k1m16n12 KERNEL_h_k1m16n12 "addq $16,%%r15;"
 #if defined(TRMMKERNEL) && !defined(LEFT) && (BACKWARDS == 0)
+ #ifdef BROADCAST_KERNEL
   #define loada_kend_k1m16 "vmovupd (%0,%3,1),%%zmm1; vmovupd 64(%0,%3,1),%%zmm2; addq $128,%3;"
-  #define acc_kend_nc2_k1m16(boff1,boff2) unit_acc_gen_m16n2(12,13,14,15,boff1,boff2,%1,%%r12,1)
-  #define acc_kend_nc3_k1m16(boff1,boff2) unit_acc_gen_m16n2(16,17,18,19,boff1,boff2,%1,%%r12,2)
-  #define acc_kend_nc4_k1m16(boff1,boff2) unit_acc_gen_m16n2(20,21,22,23,boff1,boff2,%%r15)
-  #define acc_kend_nc5_k1m16(boff1,boff2) unit_acc_gen_m16n2(24,25,26,27,boff1,boff2,%%r15,%%r12,1)
-  #define acc_kend_nc6_k1m16(boff1,boff2) unit_acc_gen_m16n2(28,29,30,31,boff1,boff2,%%r15,%%r12,2)
+ #else
+  #define loada_kend_k1m16 "vmovddup (%0,%3,1),%%zmm1; vmovddup 8(%0,%3,1),%%zmm2; vmovddup 64(%0,%3,1),%%zmm3; vmovddup 72(%0,%3,1),%%zmm4; addq $128,%3;"
+ #endif
+ #define acc_kend_nc2_k1m16(boff1) unit_acc_gen_m16n2(12,13,14,15,boff1,%1,%%r12,1)
+ #define acc_kend_nc3_k1m16(boff1) unit_acc_gen_m16n2(16,17,18,19,boff1,%1,%%r12,2)
+ #define acc_kend_nc4_k1m16(boff1) unit_acc_gen_m16n2(20,21,22,23,boff1,%%r15)
+ #define acc_kend_nc5_k1m16(boff1) unit_acc_gen_m16n2(24,25,26,27,boff1,%%r15,%%r12,1)
+ #define acc_kend_nc6_k1m16(boff1) unit_acc_gen_m16n2(28,29,30,31,boff1,%%r15,%%r12,2)
 #endif
 #define save_init_m16 "movq %2,%3; addq $128,%2;"
 #ifdef TRMMKERNEL
   #define SAVE_m16n1 "vmulpd %%zmm8,%%zmm0,%%zmm8; vmovupd %%zmm8,(%2); vmulpd %%zmm9,%%zmm0,%%zmm9; vmovupd %%zmm9,64(%2); addq $128,%2;"
+ #ifdef BROADCAST_KERNEL
   #define unit_save_m16n2(c1_no,c2_no,c3_no,c4_no)\
     "vmulpd %%zmm"#c1_no",%%zmm0,%%zmm"#c1_no"; vmovupd %%zmm"#c1_no",(%3); vmulpd %%zmm"#c2_no",%%zmm0,%%zmm"#c2_no"; vmovupd %%zmm"#c2_no",64(%3);"\
     "vmulpd %%zmm"#c3_no",%%zmm0,%%zmm"#c3_no"; vmovupd %%zmm"#c3_no",(%3,%4,1); vmulpd %%zmm"#c4_no",%%zmm0,%%zmm"#c4_no"; vmovupd %%zmm"#c4_no",64(%3,%4,1); leaq (%3,%4,2),%3;"
+ #else
+  #define unit_save_m16n2(c1_no,c2_no,c3_no,c4_no)\
+    "vunpcklpd %%zmm"#c2_no",%%zmm"#c1_no",%%zmm1; vunpcklpd %%zmm"#c4_no",%%zmm"#c3_no",%%zmm2; vunpckhpd %%zmm"#c2_no",%%zmm"#c1_no",%%zmm3; vunpckhpd %%zmm"#c4_no",%%zmm"#c3_no",%%zmm4;"\
+    "vmulpd %%zmm1,%%zmm0,%%zmm1; vmovupd %%zmm1,(%3); vmulpd %%zmm2,%%zmm0,%%zmm2; vmovupd %%zmm2,64(%3);"\
+    "vmulpd %%zmm3,%%zmm0,%%zmm3; vmovupd %%zmm3,(%3,%4,1); vmulpd %%zmm4,%%zmm0,%%zmm4; vmovupd %%zmm4,64(%3,%4,1); leaq (%3,%4,2),%3;"
+ #endif
 #else
   #define SAVE_m16n1 "vfmadd213pd (%2),%%zmm0,%%zmm8; vmovupd %%zmm8,(%2); vfmadd213pd 64(%2),%%zmm0,%%zmm9; vmovupd %%zmm9,64(%2); addq $128,%2;"
+ #ifdef BROADCAST_KERNEL
   #define unit_save_m16n2(c1_no,c2_no,c3_no,c4_no)\
     "vfmadd213pd (%3),%%zmm0,%%zmm"#c1_no"; vmovupd %%zmm"#c1_no",(%3); vfmadd213pd 64(%3),%%zmm0,%%zmm"#c2_no"; vmovupd %%zmm"#c2_no",64(%3);"\
     "vfmadd213pd (%3,%4,1),%%zmm0,%%zmm"#c3_no"; vmovupd %%zmm"#c3_no",(%3,%4,1); vfmadd213pd 64(%3,%4,1),%%zmm0,%%zmm"#c4_no"; vmovupd %%zmm"#c4_no",64(%3,%4,1); leaq (%3,%4,2),%3;"
+ #else
+  #define unit_save_m16n2(c1_no,c2_no,c3_no,c4_no)\
+    "vunpcklpd %%zmm"#c2_no",%%zmm"#c1_no",%%zmm1; vunpcklpd %%zmm"#c4_no",%%zmm"#c3_no",%%zmm2; vunpckhpd %%zmm"#c2_no",%%zmm"#c1_no",%%zmm3; vunpckhpd %%zmm"#c4_no",%%zmm"#c3_no",%%zmm4;"\
+    "vfmadd213pd (%3),%%zmm0,%%zmm1; vmovupd %%zmm1,(%3); vfmadd213pd 64(%3),%%zmm0,%%zmm2; vmovupd %%zmm2,64(%3);"\
+    "vfmadd213pd (%3,%4,1),%%zmm0,%%zmm3; vmovupd %%zmm3,(%3,%4,1); vfmadd213pd 64(%3,%4,1),%%zmm0,%%zmm4; vmovupd %%zmm4,64(%3,%4,1); leaq (%3,%4,2),%3;"
+ #endif
 #endif
 #define SAVE_m16n2 save_init_m16 unit_save_m16n2(8,9,10,11)
 #define SAVE_m16n4 SAVE_m16n2 unit_save_m16n2(12,13,14,15)
@@ -206,11 +234,11 @@
 #define KERNEL_k1m8n12 KERNEL_h_k1m8n12 "addq $16,%%r15;"
 #if defined(TRMMKERNEL) && !defined(LEFT) && (BACKWARDS == 0)
   #define loada_kend_k1m8 "vmovddup (%0,%3,1),%%zmm1; vmovddup 8(%0,%3,1),%%zmm2; addq $64,%3;"
-  #define acc_kend_nc2_k1m8(boff1,boff2) unit_acc_gen_m8n2(10,11,boff1,%1,%%r12,1)
-  #define acc_kend_nc3_k1m8(boff1,boff2) unit_acc_gen_m8n2(12,13,boff1,%1,%%r12,2)
-  #define acc_kend_nc4_k1m8(boff1,boff2) unit_acc_gen_m8n2(14,15,boff1,%%r15)
-  #define acc_kend_nc5_k1m8(boff1,boff2) unit_acc_gen_m8n2(16,17,boff1,%%r15,%%r12,1)
-  #define acc_kend_nc6_k1m8(boff1,boff2) unit_acc_gen_m8n2(18,19,boff1,%%r15,%%r12,2)
+  #define acc_kend_nc2_k1m8(boff1) unit_acc_gen_m8n2(10,11,boff1,%1,%%r12,1)
+  #define acc_kend_nc3_k1m8(boff1) unit_acc_gen_m8n2(12,13,boff1,%1,%%r12,2)
+  #define acc_kend_nc4_k1m8(boff1) unit_acc_gen_m8n2(14,15,boff1,%%r15)
+  #define acc_kend_nc5_k1m8(boff1) unit_acc_gen_m8n2(16,17,boff1,%%r15,%%r12,1)
+  #define acc_kend_nc6_k1m8(boff1) unit_acc_gen_m8n2(18,19,boff1,%%r15,%%r12,2)
 #endif
 #define save_init_m8 "movq %2,%3; addq $64,%2;"
 #ifdef TRMMKERNEL
@@ -258,11 +286,11 @@
 #define KERNEL_k1m4n12 KERNEL_h_k1m4n12 "addq $16,%%r15;"
 #if defined(TRMMKERNEL) && !defined(LEFT) && (BACKWARDS == 0)
   #define loada_kend_k1m4 "vmovddup (%0,%3,1),%%ymm1; vmovddup 8(%0,%3,1),%%ymm2; addq $32,%3;"
-  #define acc_kend_nc2_k1m4(boff1,boff2) unit_acc_gen_m4n2(6,7,boff1,%1,%%r12,1)
-  #define acc_kend_nc3_k1m4(boff1,boff2) unit_acc_gen_m4n2(8,9,boff1,%1,%%r12,2)
-  #define acc_kend_nc4_k1m4(boff1,boff2) unit_acc_gen_m4n2(10,11,boff1,%%r15)
-  #define acc_kend_nc5_k1m4(boff1,boff2) unit_acc_gen_m4n2(12,13,boff1,%%r15,%%r12,1)
-  #define acc_kend_nc6_k1m4(boff1,boff2) unit_acc_gen_m4n2(14,15,boff1,%%r15,%%r12,2)
+  #define acc_kend_nc2_k1m4(boff1) unit_acc_gen_m4n2(6,7,boff1,%1,%%r12,1)
+  #define acc_kend_nc3_k1m4(boff1) unit_acc_gen_m4n2(8,9,boff1,%1,%%r12,2)
+  #define acc_kend_nc4_k1m4(boff1) unit_acc_gen_m4n2(10,11,boff1,%%r15)
+  #define acc_kend_nc5_k1m4(boff1) unit_acc_gen_m4n2(12,13,boff1,%%r15,%%r12,1)
+  #define acc_kend_nc6_k1m4(boff1) unit_acc_gen_m4n2(14,15,boff1,%%r15,%%r12,2)
 #endif
 #define save_init_m4 "movq %2,%3; addq $32,%2;"
 #ifdef TRMMKERNEL
@@ -311,11 +339,11 @@
 #define KERNEL_k1m2n12 KERNEL_h_k1m2n12 "addq $16,%%r15;"
 #if defined(TRMMKERNEL) && !defined(LEFT) && (BACKWARDS == 0)
   #define loada_kend_k1m2 "vmovddup (%0,%3,1),%%xmm1; vmovddup 8(%0,%3,1),%%xmm2; addq $16,%3;"
-  #define acc_kend_nc2_k1m2(boff1,boff2) unit_acc_gen_m2n2(6,7,boff1,%1,%%r12,1)
-  #define acc_kend_nc3_k1m2(boff1,boff2) unit_acc_gen_m2n2(8,9,boff1,%1,%%r12,2)
-  #define acc_kend_nc4_k1m2(boff1,boff2) unit_acc_gen_m2n2(10,11,boff1,%%r15)
-  #define acc_kend_nc5_k1m2(boff1,boff2) unit_acc_gen_m2n2(12,13,boff1,%%r15,%%r12,1)
-  #define acc_kend_nc6_k1m2(boff1,boff2) unit_acc_gen_m2n2(14,15,boff1,%%r15,%%r12,2)
+  #define acc_kend_nc2_k1m2(boff1) unit_acc_gen_m2n2(6,7,boff1,%1,%%r12,1)
+  #define acc_kend_nc3_k1m2(boff1) unit_acc_gen_m2n2(8,9,boff1,%1,%%r12,2)
+  #define acc_kend_nc4_k1m2(boff1) unit_acc_gen_m2n2(10,11,boff1,%%r15)
+  #define acc_kend_nc5_k1m2(boff1) unit_acc_gen_m2n2(12,13,boff1,%%r15,%%r12,1)
+  #define acc_kend_nc6_k1m2(boff1) unit_acc_gen_m2n2(14,15,boff1,%%r15,%%r12,2)
 #endif
 #define save_init_m2 "movq %2,%3; addq $16,%2;"
 #ifdef TRMMKERNEL
@@ -362,11 +390,11 @@
 #define KERNEL_k1m1n12 KERNEL_h_k1m1n12 "addq $16,%%r15;"
 #if defined(TRMMKERNEL) && !defined(LEFT) && (BACKWARDS == 0)
   #define loada_kend_k1m1 "vmovddup (%0,%3,1),%%xmm1; addq $8,%3;"
-  #define acc_kend_nc2_k1m1(boff1,boff2) "vfmadd231pd "#boff1"(%1,%%r12,1),%%xmm1,%%xmm5;"
-  #define acc_kend_nc3_k1m1(boff1,boff2) "vfmadd231pd "#boff1"(%1,%%r12,2),%%xmm1,%%xmm6;"
-  #define acc_kend_nc4_k1m1(boff1,boff2) "vfmadd231pd "#boff1"(%%r15),%%xmm1,%%xmm7;"
-  #define acc_kend_nc5_k1m1(boff1,boff2) "vfmadd231pd "#boff1"(%%r15,%%r12,1),%%xmm1,%%xmm8;"
-  #define acc_kend_nc6_k1m1(boff1,boff2) "vfmadd231pd "#boff1"(%%r15,%%r12,2),%%xmm1,%%xmm9;"
+  #define acc_kend_nc2_k1m1(boff1) "vfmadd231pd "#boff1"(%1,%%r12,1),%%xmm1,%%xmm5;"
+  #define acc_kend_nc3_k1m1(boff1) "vfmadd231pd "#boff1"(%1,%%r12,2),%%xmm1,%%xmm6;"
+  #define acc_kend_nc4_k1m1(boff1) "vfmadd231pd "#boff1"(%%r15),%%xmm1,%%xmm7;"
+  #define acc_kend_nc5_k1m1(boff1) "vfmadd231pd "#boff1"(%%r15,%%r12,1),%%xmm1,%%xmm8;"
+  #define acc_kend_nc6_k1m1(boff1) "vfmadd231pd "#boff1"(%%r15,%%r12,2),%%xmm1,%%xmm9;"
 #endif
 #define save_init_m1 "movq %2,%3; addq $8,%2;"
 #ifdef TRMMKERNEL
diff --git a/kernel/x86_64/dsymv_L_microk_skylakex-2.c b/kernel/x86_64/dsymv_L_microk_skylakex-2.c
index bdcd914fb..f0df5aaa8 100644
--- a/kernel/x86_64/dsymv_L_microk_skylakex-2.c
+++ b/kernel/x86_64/dsymv_L_microk_skylakex-2.c
@@ -36,7 +36,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #if defined(__clang_patchlevel__) && __clang_major__ == 9 && __clang_minor__ == 0 && __clang_patchlevel__ == 0
 #pragma clang optimize off
 #endif
-
+#if defined(__apple_build_version__) && __clang_major__ == 11 && __clang_minor__ == 0 && __clang_patchlevel__ == 3
+#pragma clang optimize off
+#endif
 static void dsymv_kernel_4x4(BLASLONG from, BLASLONG to, FLOAT **a, FLOAT *x, FLOAT *y, FLOAT *temp1, FLOAT *temp2)
 {
 
@@ -164,6 +166,9 @@ static void dsymv_kernel_4x4(BLASLONG from, BLASLONG to, FLOAT **a, FLOAT *x, FL
 #if defined(__clang_patchlevel__) && __clang_major__ == 9 && __clang_minor__ == 0 && __clang_patchlevel__ == 0
 #pragma clang optimize on
 #endif
+#if defined(__apple_build_version__) && __clang_major__ == 11 && __clang_minor__ == 0 && __clang_patchlevel__ == 3
+#pragma clang optimize on
+#endif
 
 #else
 #include "dsymv_L_microk_haswell-2.c"
diff --git a/kernel/x86_64/sgemm_kernel_16x4_skylakex_3.c b/kernel/x86_64/sgemm_kernel_16x4_skylakex_3.c
new file mode 100644
index 000000000..3b1af33c1
--- /dev/null
+++ b/kernel/x86_64/sgemm_kernel_16x4_skylakex_3.c
@@ -0,0 +1,515 @@
+/* %0 = "+r"(a_pointer), %1 = "+r"(b_pointer), %2 = "+r"(c_pointer), %3 = "+r"(ldc_in_bytes), %4 for k_count, %5 for c_store */
+/* r10 to assist prefetch, r12 = k << 4(const), r13 = k(const), r14 = b_head_pos(const), r15 = %1 + 3r12 */
+
+#include "common.h"
+#include <stdint.h>
+
+/* m = 16 */ /* zmm8-zmm31 for accumulators, zmm4-zmm7 for temporary use, zmm0 for alpha */
+#define KERNEL_k1m16n1 \
+    "vmovups (%0),%%zmm4; addq $64,%0;"\
+    "vbroadcastss (%1),%%zmm6; vfmadd231ps %%zmm4,%%zmm6,%%zmm8;"\
+    "addq $4,%1;"
+#define KERNEL_h_k1m16n2 \
+    "vmovsldup (%0),%%zmm4; vmovshdup (%0),%%zmm5; prefetcht0 512(%0); addq $64,%0;"\
+    "vbroadcastsd (%1),%%zmm6; vfmadd231ps %%zmm4,%%zmm6,%%zmm8; vfmadd231ps %%zmm5,%%zmm6,%%zmm9;"
+#define KERNEL_k1m16n2 KERNEL_h_k1m16n2 "addq $8,%1;"
+#define KERNEL_h_k1m16n4 KERNEL_h_k1m16n2 "vbroadcastsd 8(%1),%%zmm7; vfmadd231ps %%zmm4,%%zmm7,%%zmm10; vfmadd231ps %%zmm5,%%zmm7,%%zmm11;"
+#define KERNEL_k1m16n4 KERNEL_h_k1m16n4 "addq $16,%1;"
+#define unit_kernel_k1m16n4(c1,c2,c3,c4, ...) \
+    "vbroadcastsd  ("#__VA_ARGS__"),%%zmm6; vfmadd231ps %%zmm4,%%zmm6,"#c1"; vfmadd231ps %%zmm5,%%zmm6,"#c2";"\
+    "vbroadcastsd 8("#__VA_ARGS__"),%%zmm7; vfmadd231ps %%zmm4,%%zmm7,"#c3"; vfmadd231ps %%zmm5,%%zmm7,"#c4";"
+#define KERNEL_h_k1m16n8 KERNEL_h_k1m16n4 unit_kernel_k1m16n4(%%zmm12,%%zmm13,%%zmm14,%%zmm15,%1,%%r12,1)
+#define KERNEL_k1m16n8 KERNEL_h_k1m16n8 "addq $16,%1;"
+#define KERNEL_h_k1m16n12 KERNEL_h_k1m16n8 unit_kernel_k1m16n4(%%zmm16,%%zmm17,%%zmm18,%%zmm19,%1,%%r12,2)
+#define KERNEL_k1m16n12 KERNEL_h_k1m16n12 "addq $16,%1;"
+#define KERNEL_h_k1m16n16 KERNEL_k1m16n12 unit_kernel_k1m16n4(%%zmm20,%%zmm21,%%zmm22,%%zmm23,%%r15)
+#define KERNEL_k1m16n16 KERNEL_h_k1m16n16 "addq $16,%%r15;"
+#define KERNEL_h_k1m16n20 KERNEL_h_k1m16n16 unit_kernel_k1m16n4(%%zmm24,%%zmm25,%%zmm26,%%zmm27,%%r15,%%r12,1)
+#define KERNEL_k1m16n20 KERNEL_h_k1m16n20 "addq $16,%%r15;"
+#define KERNEL_h_k1m16n24 KERNEL_h_k1m16n20 unit_kernel_k1m16n4(%%zmm28,%%zmm29,%%zmm30,%%zmm31,%%r15,%%r12,2)
+#define KERNEL_k1m16n24 KERNEL_h_k1m16n24 "addq $16,%%r15;"
+#define INIT_m16n1 "vpxorq %%zmm8,%%zmm8,%%zmm8;"
+#define INIT_m16n2 INIT_m16n1 "vpxorq %%zmm9,%%zmm9,%%zmm9;"
+#define INIT_m16n4 INIT_m16n2 "vpxorq %%zmm10,%%zmm10,%%zmm10;vpxorq %%zmm11,%%zmm11,%%zmm11;"
+#define unit_init_m16n4(c1,c2,c3,c4) \
+    "vpxorq "#c1","#c1","#c1";vpxorq "#c2","#c2","#c2";vpxorq "#c3","#c3","#c3";vpxorq "#c4","#c4","#c4";"
+#define INIT_m16n8 INIT_m16n4 unit_init_m16n4(%%zmm12,%%zmm13,%%zmm14,%%zmm15)
+#define INIT_m16n12 INIT_m16n8 unit_init_m16n4(%%zmm16,%%zmm17,%%zmm18,%%zmm19)
+#define INIT_m16n16 INIT_m16n12 unit_init_m16n4(%%zmm20,%%zmm21,%%zmm22,%%zmm23)
+#define INIT_m16n20 INIT_m16n16 unit_init_m16n4(%%zmm24,%%zmm25,%%zmm26,%%zmm27)
+#define INIT_m16n24 INIT_m16n20 unit_init_m16n4(%%zmm28,%%zmm29,%%zmm30,%%zmm31)
+#define SAVE_h_m16n1 "vfmadd213ps (%2),%%zmm0,%%zmm8; vmovups %%zmm8,(%2);"
+#define unit_save_m16n2(c1,c2) \
+    "vunpcklps "#c2","#c1",%%zmm6; vunpckhps "#c2","#c1",%%zmm7; vunpcklpd %%zmm7,%%zmm6,%%zmm4; vunpckhpd %%zmm7,%%zmm6,%%zmm5;"\
+    "vfmadd213ps (%5),%%zmm0,%%zmm4; vfmadd213ps (%5,%3,1),%%zmm0,%%zmm5;"\
+    "vmovups %%zmm4,(%5); vmovups %%zmm5,(%5,%3,1); leaq (%5,%3,2),%5;"
+#define SAVE_h_m16n2 "movq %2,%5;" unit_save_m16n2(%%zmm8,%%zmm9)
+#define SAVE_h_m16n4  SAVE_h_m16n2  unit_save_m16n2(%%zmm10,%%zmm11)
+#define SAVE_h_m16n8  SAVE_h_m16n4  unit_save_m16n2(%%zmm12,%%zmm13) unit_save_m16n2(%%zmm14,%%zmm15)
+#define SAVE_h_m16n12 SAVE_h_m16n8  unit_save_m16n2(%%zmm16,%%zmm17) unit_save_m16n2(%%zmm18,%%zmm19)
+#define SAVE_h_m16n16 SAVE_h_m16n12 unit_save_m16n2(%%zmm20,%%zmm21) unit_save_m16n2(%%zmm22,%%zmm23)
+#define SAVE_h_m16n20 SAVE_h_m16n16 unit_save_m16n2(%%zmm24,%%zmm25) unit_save_m16n2(%%zmm26,%%zmm27)
+#define SAVE_h_m16n24 SAVE_h_m16n20 unit_save_m16n2(%%zmm28,%%zmm29) unit_save_m16n2(%%zmm30,%%zmm31)
+#define SAVE_m16(ndim) SAVE_h_m16n##ndim "addq $64,%2;"
+#define COMPUTE_m16(ndim) \
+    INIT_m16n##ndim\
+    "movq %%r13,%4; movq %%r14,%1; leaq (%1,%%r12,2),%%r15; addq %%r12,%%r15; movq %2,%5; xorq %%r10,%%r10;"\
+    "cmpq $16,%4; jb "#ndim"016162f;"\
+    #ndim"016161:\n\t"\
+    "cmpq $126,%%r10; movq $126,%%r10; cmoveq %3,%%r10;"\
+    KERNEL_k1m16n##ndim\
+    KERNEL_k1m16n##ndim\
+    "prefetcht1 (%5); subq $63,%5; addq %%r10,%5;"\
+    KERNEL_k1m16n##ndim\
+    KERNEL_k1m16n##ndim\
+    "prefetcht1 (%6); addq $32,%6;"\
+    "subq $4,%4; cmpq $16,%4; jnb "#ndim"016161b;"\
+    "movq %2,%5;"\
+    #ndim"016162:\n\t"\
+    "testq %4,%4; jz "#ndim"016164f;"\
+    #ndim"016163:\n\t"\
+    "prefetcht0 (%5); prefetcht0 63(%5); prefetcht0 (%5,%3,1); prefetcht0 63(%5,%3,1);"\
+    KERNEL_k1m16n##ndim\
+    "leaq (%5,%3,2),%5; decq %4; jnz "#ndim"016163b;"\
+    #ndim"016164:\n\t"\
+    "prefetcht0 (%%r14); prefetcht0 64(%%r14);"\
+    SAVE_m16(ndim)
+#define unit_save_m16n2_rscr(c1,c2,scr_off) \
+    "vunpcklps "#c2","#c1",%%zmm6; vunpckhps "#c2","#c1",%%zmm7; vunpcklpd %%zmm7,%%zmm6,%%zmm4; vunpckhpd %%zmm7,%%zmm6,%%zmm5;"\
+    "vmovups "#scr_off"(%7),%%zmm6; vfmadd213ps -64(%5),%%zmm0,%%zmm6; vfmadd213ps (%5),%%zmm0,%%zmm4;"\
+    "vmovups %%zmm6,-64(%5); vmovups %%zmm4,(%5);"\
+    "vmovups "#scr_off"+64(%7),%%zmm6; vfmadd213ps -64(%5,%3,1),%%zmm0,%%zmm6; vfmadd213ps (%5,%3,1),%%zmm0,%%zmm5;"\
+    "vmovups %%zmm6,-64(%5,%3,1); vmovups %%zmm5,(%5,%3,1); leaq (%5,%3,2),%5;"
+#define unit_save_m16n2_wscr(c1,c2,scr_off) \
+    "vunpcklps "#c2","#c1",%%zmm6; vunpckhps "#c2","#c1",%%zmm7; vunpcklpd %%zmm7,%%zmm6,%%zmm4; vunpckhpd %%zmm7,%%zmm6,%%zmm5;"\
+    "vmovups %%zmm4,"#scr_off"(%7); vmovups %%zmm5,"#scr_off"+64(%7);"
+#define COMPUTE_m16n24_LSAVE \
+    INIT_m16n24\
+    "movq %%r13,%4; movq %%r14,%1; leaq (%1,%%r12,2),%%r15; addq %%r12,%%r15; movq %2,%5;"\
+    "cmpq $16,%4; jb 24716162f; movq $16,%4;"\
+    "24716161:\n\t"\
+    KERNEL_k1m16n24 "addq $4,%4; testq $12,%4; movq $172,%%r10; cmovz %3,%%r10;"\
+    KERNEL_k1m16n24 "prefetcht1 -64(%5); leaq -129(%5,%%r10,1),%5;"\
+    KERNEL_k1m16n24 "prefetcht1 (%6); addq $32,%6; cmpq $208,%4; cmoveq %2,%5;"\
+    KERNEL_k1m16n24 "cmpq %4,%%r13; jnb 24716161b;"\
+    "movq %2,%5; negq %4; leaq 16(%%r13,%4,1),%4;"\
+    "24716162:\n\t"\
+    "testq %4,%4; jz 24716164f; movq %7,%%r10;"\
+    "24716163:\n\t"\
+    "prefetcht0 -64(%5); prefetcht0 (%5); prefetcht0 63(%5); addq %3,%5;"\
+    KERNEL_k1m16n24 "prefetcht0 (%%r10); addq $64,%%r10; decq %4; jnz 24716163b;"\
+    "24716164:\n\t"\
+    "prefetcht0 (%%r14); prefetcht0 64(%%r14); movq %2,%5; addq $64,%2;"\
+    unit_save_m16n2_rscr(%%zmm8,%%zmm9,0) unit_save_m16n2_rscr(%%zmm10,%%zmm11,128) unit_save_m16n2_rscr(%%zmm12,%%zmm13,256)\
+    unit_save_m16n2_rscr(%%zmm14,%%zmm15,384) unit_save_m16n2_rscr(%%zmm16,%%zmm17,512) unit_save_m16n2_rscr(%%zmm18,%%zmm19,640)\
+    unit_save_m16n2_wscr(%%zmm20,%%zmm21,0) unit_save_m16n2_wscr(%%zmm22,%%zmm23,128) unit_save_m16n2_wscr(%%zmm24,%%zmm25,256)\
+    unit_save_m16n2_wscr(%%zmm26,%%zmm27,384) unit_save_m16n2_wscr(%%zmm28,%%zmm29,512) unit_save_m16n2_wscr(%%zmm30,%%zmm31,640)
+#define COMPUTE_m16n24_RSAVE \
+    INIT_m16n24 "leaq (%2,%3,8),%2; leaq (%2,%3,4),%2;"\
+    "movq %%r13,%4; movq %%r14,%1; leaq (%1,%%r12,2),%%r15; addq %%r12,%%r15; movq %2,%5;"\
+    "cmpq $16,%4; jb 24616162f; movq $16,%4;"\
+    "24616161:\n\t"\
+    KERNEL_k1m16n24 "addq $4,%4; testq $12,%4; movq $172,%%r10; cmovz %3,%%r10;"\
+    KERNEL_k1m16n24 "prefetcht1 -64(%5); leaq -129(%5,%%r10,1),%5;"\
+    KERNEL_k1m16n24 "prefetcht1 (%6); addq $32,%6; cmpq $208,%4; cmoveq %2,%5;"\
+    KERNEL_k1m16n24 "cmpq %4,%%r13; jnb 24616161b;"\
+    "movq %2,%5; negq %4; leaq 16(%%r13,%4,1),%4;"\
+    "24616162:\n\t"\
+    "testq %4,%4; jz 24616164f; movq %7,%%r10;"\
+    "24616163:\n\t"\
+    "prefetcht0 -64(%5); prefetcht0 (%5); prefetcht0 63(%5); addq %3,%5;"\
+    KERNEL_k1m16n24 "prefetcht0 (%%r10); addq $64,%%r10; decq %4; jnz 24616163b;"\
+    "24616164:\n\t"\
+    "prefetcht0 (%%r14); prefetcht0 64(%%r14); movq %2,%5; addq $64,%2;"\
+    unit_save_m16n2_rscr(%%zmm20,%%zmm21,0) unit_save_m16n2_rscr(%%zmm22,%%zmm23,128) unit_save_m16n2_rscr(%%zmm24,%%zmm25,256)\
+    unit_save_m16n2_rscr(%%zmm26,%%zmm27,384) unit_save_m16n2_rscr(%%zmm28,%%zmm29,512) unit_save_m16n2_rscr(%%zmm30,%%zmm31,640)\
+    unit_save_m16n2_wscr(%%zmm8,%%zmm9,0) unit_save_m16n2_wscr(%%zmm10,%%zmm11,128) unit_save_m16n2_wscr(%%zmm12,%%zmm13,256)\
+    unit_save_m16n2_wscr(%%zmm14,%%zmm15,384) unit_save_m16n2_wscr(%%zmm16,%%zmm17,512) unit_save_m16n2_wscr(%%zmm18,%%zmm19,640)\
+    "negq %3; leaq (%2,%3,8),%2; leaq (%2,%3,4),%2; negq %3;"
+#define COMPUTE_m16n24_LINIT \
+    INIT_m16n24\
+    "movq %%r13,%4; movq %%r14,%1; leaq (%1,%%r12,2),%%r15; addq %%r12,%%r15; movq %2,%5;"\
+    "cmpq $16,%4; jb 24516162f; movq $16,%4;"\
+    "24516161:\n\t"\
+    KERNEL_k1m16n24 "addq $4,%4; testq $12,%4; movq $84,%%r10; cmovz %3,%%r10;"\
+    KERNEL_k1m16n24 "prefetcht1 (%5); leaq -63(%5,%%r10,1),%5;"\
+    KERNEL_k1m16n24 "prefetcht1 (%6); addq $32,%6; cmpq $208,%4; cmoveq %2,%5;"\
+    KERNEL_k1m16n24 "cmpq %4,%%r13; jnb 24516161b;"\
+    "movq %2,%5; negq %4; leaq 16(%%r13,%4,1),%4;"\
+    "24516162:\n\t"\
+    "testq %4,%4; jz 24516164f; movq %7,%%r10;"\
+    "24516163:\n\t"\
+    "prefetcht0 (%5); prefetcht0 63(%5); addq %3,%5;"\
+    KERNEL_k1m16n24 "prefetcht0 (%%r10); addq $64,%%r10; decq %4; jnz 24516163b;"\
+    "24516164:\n\t"\
+    "prefetcht0 (%%r14); prefetcht0 64(%%r14); movq %2,%5; addq $64,%2;"\
+    unit_save_m16n2(%%zmm8,%%zmm9) unit_save_m16n2(%%zmm10,%%zmm11) unit_save_m16n2(%%zmm12,%%zmm13)\
+    unit_save_m16n2(%%zmm14,%%zmm15) unit_save_m16n2(%%zmm16,%%zmm17) unit_save_m16n2(%%zmm18,%%zmm19)\
+    unit_save_m16n2_wscr(%%zmm20,%%zmm21,0) unit_save_m16n2_wscr(%%zmm22,%%zmm23,128) unit_save_m16n2_wscr(%%zmm24,%%zmm25,256)\
+    unit_save_m16n2_wscr(%%zmm26,%%zmm27,384) unit_save_m16n2_wscr(%%zmm28,%%zmm29,512) unit_save_m16n2_wscr(%%zmm30,%%zmm31,640)
+#define COMPUTE_m16n24_LTAIL \
+    INIT_m16n24\
+    "movq %%r13,%4; movq %%r14,%1; leaq (%1,%%r12,2),%%r15; addq %%r12,%%r15; movq %2,%5;"\
+    "cmpq $16,%4; jb 24416162f; movq $16,%4;"\
+    "24416161:\n\t"\
+    KERNEL_k1m16n24 "addq $4,%4; testq $4,%4; movq $126,%%r10; cmovz %3,%%r10;"\
+    KERNEL_k1m16n24 "prefetcht1 -64(%5); prefetcht1 (%5); leaq -63(%5,%%r10,1),%5;"\
+    KERNEL_k1m16n24 "prefetcht1 (%6); addq $32,%6; cmpq $208,%4; cmoveq %2,%5;"\
+    KERNEL_k1m16n24 "cmpq %4,%%r13; jnb 24416161b;"\
+    "movq %2,%5; negq %4; leaq 16(%%r13,%4,1),%4;"\
+    "24416162:\n\t"\
+    "testq %4,%4; jz 24416164f; movq %7,%%r10;"\
+    "24416163:\n\t"\
+    "prefetcht0 -64(%5); prefetcht0 (%5); prefetcht0 63(%5); prefetcht0 -64(%5,%3,1); prefetcht0 (%5,%3,1); prefetcht0 63(%5,%3,1); leaq (%5,%3,2),%5;"\
+    KERNEL_k1m16n24 "prefetcht0 (%%r10); addq $64,%%r10; decq %4; jnz 24416163b;"\
+    "24416164:\n\t"\
+    "prefetcht0 (%%r14); prefetcht0 64(%%r14); movq %2,%5; addq $64,%2;"\
+    unit_save_m16n2_rscr(%%zmm8,%%zmm9,0) unit_save_m16n2_rscr(%%zmm10,%%zmm11,128) unit_save_m16n2_rscr(%%zmm12,%%zmm13,256)\
+    unit_save_m16n2_rscr(%%zmm14,%%zmm15,384) unit_save_m16n2_rscr(%%zmm16,%%zmm17,512) unit_save_m16n2_rscr(%%zmm18,%%zmm19,640)\
+    unit_save_m16n2(%%zmm20,%%zmm21) unit_save_m16n2(%%zmm22,%%zmm23) unit_save_m16n2(%%zmm24,%%zmm25)\
+    unit_save_m16n2(%%zmm26,%%zmm27) unit_save_m16n2(%%zmm28,%%zmm29) unit_save_m16n2(%%zmm30,%%zmm31)
+#define COMPUTE_m16n24_RTAIL \
+    INIT_m16n24\
+    "movq %%r13,%4; movq %%r14,%1; leaq (%1,%%r12,2),%%r15; addq %%r12,%%r15; movq %2,%5;"\
+    "cmpq $16,%4; jb 24416162f; movq $16,%4;"\
+    "24416161:\n\t"\
+    KERNEL_k1m16n24 "addq $4,%4; testq $4,%4; movq $126,%%r10; cmovz %3,%%r10;"\
+    KERNEL_k1m16n24 "prefetcht1 -64(%5); prefetcht1 (%5); leaq -63(%5,%%r10,1),%5;"\
+    KERNEL_k1m16n24 "prefetcht1 (%6); addq $32,%6; cmpq $208,%4; cmoveq %2,%5;"\
+    KERNEL_k1m16n24 "cmpq %4,%%r13; jnb 24416161b;"\
+    "movq %2,%5; negq %4; leaq 16(%%r13,%4,1),%4;"\
+    "24416162:\n\t"\
+    "testq %4,%4; jz 24416164f; movq %7,%%r10;"\
+    "24416163:\n\t"\
+    "prefetcht0 -64(%5); prefetcht0 (%5); prefetcht0 63(%5); prefetcht0 -64(%5,%3,1); prefetcht0 (%5,%3,1); prefetcht0 63(%5,%3,1); leaq (%5,%3,2),%5;"\
+    KERNEL_k1m16n24 "prefetcht0 (%%r10); addq $64,%%r10; decq %4; jnz 24416163b;"\
+    "24416164:\n\t"\
+    "prefetcht0 (%%r14); prefetcht0 64(%%r14); movq %2,%5; addq $64,%2;"\
+    unit_save_m16n2(%%zmm8,%%zmm9) unit_save_m16n2(%%zmm10,%%zmm11) unit_save_m16n2(%%zmm12,%%zmm13)\
+    unit_save_m16n2(%%zmm14,%%zmm15) unit_save_m16n2(%%zmm16,%%zmm17) unit_save_m16n2(%%zmm18,%%zmm19)\
+    unit_save_m16n2_rscr(%%zmm20,%%zmm21,0) unit_save_m16n2_rscr(%%zmm22,%%zmm23,128) unit_save_m16n2_rscr(%%zmm24,%%zmm25,256)\
+    unit_save_m16n2_rscr(%%zmm26,%%zmm27,384) unit_save_m16n2_rscr(%%zmm28,%%zmm29,512) unit_save_m16n2_rscr(%%zmm30,%%zmm31,640)
+
+/* m = 8 *//* zmm0 for alpha, zmm1-2 for perm words, zmm4-7 for temporary use, zmm8-19 for accumulators */
+#define KERNEL_k1m8n1 \
+    "vbroadcastss (%1),%%ymm4; addq $4,%1; vfmadd231ps (%0),%%ymm4,%%ymm8; addq $32,%0;"
+#define KERNEL_k1m8n2 \
+    "vmovups (%0),%%ymm4; addq $32,%0;"\
+    "vbroadcastss (%1),%%ymm5; vfmadd231ps %%ymm5,%%ymm4,%%ymm8;"\
+    "vbroadcastss 4(%1),%%ymm6; vfmadd231ps %%ymm6,%%ymm4,%%ymm9; addq $8,%1;"
+#define unit_kernel_k1m8n4(c1,c2,...)\
+    "vbroadcastf32x4 ("#__VA_ARGS__"),%%zmm7; vfmadd231ps %%zmm7,%%zmm4,"#c1"; vfmadd231ps %%zmm7,%%zmm5,"#c2";"
+#define KERNEL_h_k1m8n4 \
+    "vbroadcastf32x4 (%0),%%zmm4; vpermilps %%zmm2,%%zmm4,%%zmm4; vbroadcastf32x4 16(%0),%%zmm5; vpermilps %%zmm2,%%zmm5,%%zmm5; addq $32,%0;"\
+    unit_kernel_k1m8n4(%%zmm8,%%zmm9,%1)
+#define KERNEL_k1m8n4 KERNEL_h_k1m8n4 "addq $16,%1;"
+#define KERNEL_h_k1m8n8 KERNEL_h_k1m8n4 unit_kernel_k1m8n4(%%zmm10,%%zmm11,%1,%%r12,1)
+#define KERNEL_k1m8n8 KERNEL_h_k1m8n8 "addq $16,%1;"
+#define KERNEL_k1m8n12 KERNEL_h_k1m8n8 unit_kernel_k1m8n4(%%zmm12,%%zmm13,%1,%%r12,2) "addq $16,%1;"
+#define KERNEL_h_k1m8n16 KERNEL_k1m8n12 unit_kernel_k1m8n4(%%zmm14,%%zmm15,%%r15)
+#define KERNEL_k1m8n16 KERNEL_h_k1m8n16 "addq $16,%%r15;"
+#define KERNEL_h_k1m8n20 KERNEL_h_k1m8n16 unit_kernel_k1m8n4(%%zmm16,%%zmm17,%%r15,%%r12,1)
+#define KERNEL_k1m8n20 KERNEL_h_k1m8n20 "addq $16,%%r15;"
+#define KERNEL_k1m8n24 KERNEL_h_k1m8n20 unit_kernel_k1m8n4(%%zmm18,%%zmm19,%%r15,%%r12,2) "addq $16,%%r15;"
+#define INIT_m8n1 "vpxor %%ymm8,%%ymm8,%%ymm8;"
+#define INIT_m8n2 "vpxor %%ymm8,%%ymm8,%%ymm8; vpxor %%ymm9,%%ymm9,%%ymm9;"
+#define unit_init_m8n4(c1,c2) "vpxorq "#c1","#c1","#c1";vpxorq "#c2","#c2","#c2";"
+#define INIT_m8n4 unit_init_m8n4(%%zmm8,%%zmm9)
+#define INIT_m8n8 INIT_m8n4 unit_init_m8n4(%%zmm10,%%zmm11)
+#define INIT_m8n12 INIT_m8n8 unit_init_m8n4(%%zmm12,%%zmm13)
+#define INIT_m8n16 INIT_m8n12 unit_init_m8n4(%%zmm14,%%zmm15)
+#define INIT_m8n20 INIT_m8n16 unit_init_m8n4(%%zmm16,%%zmm17)
+#define INIT_m8n24 INIT_m8n20 unit_init_m8n4(%%zmm18,%%zmm19)
+#define SAVE_h_m8n1 "vfmadd213ps (%2),%%ymm0,%%ymm8; vmovups %%ymm8,(%2);"
+#define SAVE_h_m8n2 \
+    "vfmadd213ps (%2),%%ymm0,%%ymm8; vmovups %%ymm8,(%2);"\
+    "vfmadd213ps (%2,%3,1),%%ymm0,%%ymm9; vmovups %%ymm9,(%2,%3,1);"
+#define unit_save_m8n4(c1_no,c2_no)\
+    "vpermps %%zmm"#c1_no",%%zmm1,%%zmm"#c1_no"; vpermps %%zmm"#c2_no",%%zmm1,%%zmm"#c2_no";"\
+    "vextractf64x4 $1,%%zmm"#c1_no",%%ymm5; vextractf64x4 $1,%%zmm"#c2_no",%%ymm6;"\
+    "vmovups (%5),%%xmm4; vinsertf128 $1,(%5,%3,1),%%ymm4,%%ymm4; vfmadd231ps %%ymm"#c1_no",%%ymm0,%%ymm4;"\
+    "vmovups %%xmm4,(%5); vextractf128 $1,%%ymm4,(%5,%3,1);"\
+    "vmovups 16(%5),%%xmm4; vinsertf128 $1,16(%5,%3,1),%%ymm4,%%ymm4; vfmadd231ps %%ymm"#c2_no",%%ymm0,%%ymm4;"\
+    "vmovups %%xmm4,16(%5); vextractf128 $1,%%ymm4,16(%5,%3,1); leaq (%5,%3,2),%5;"\
+    "vmovups (%5),%%xmm4; vinsertf128 $1,(%5,%3,1),%%ymm4,%%ymm4; vfmadd231ps %%ymm5,%%ymm0,%%ymm4;"\
+    "vmovups %%xmm4,(%5); vextractf128 $1,%%ymm4,(%5,%3,1);"\
+    "vmovups 16(%5),%%xmm4; vinsertf128 $1,16(%5,%3,1),%%ymm4,%%ymm4; vfmadd231ps %%ymm6,%%ymm0,%%ymm4;"\
+    "vmovups %%xmm4,16(%5); vextractf128 $1,%%ymm4,16(%5,%3,1); leaq (%5,%3,2),%5;"
+#define SAVE_h_m8n4 "movq %2,%5;" unit_save_m8n4(8,9)
+#define SAVE_h_m8n8 SAVE_h_m8n4 unit_save_m8n4(10,11)
+#define SAVE_h_m8n12 SAVE_h_m8n8 unit_save_m8n4(12,13)
+#define SAVE_h_m8n16 SAVE_h_m8n12 unit_save_m8n4(14,15)
+#define SAVE_h_m8n20 SAVE_h_m8n16 unit_save_m8n4(16,17)
+#define SAVE_h_m8n24 SAVE_h_m8n20 unit_save_m8n4(18,19)
+#define SAVE_m8(ndim) SAVE_h_m8n##ndim "addq $32,%2;"
+#define COMPUTE_m8(ndim) \
+    INIT_m8n##ndim\
+    "movq %%r13,%4; movq %%r14,%1; leaq (%1,%%r12,2),%%r15; addq %%r12,%%r15;"\
+    "testq %4,%4; jz "#ndim"008082f;"\
+    #ndim"008081:\n\t"\
+    KERNEL_k1m8n##ndim "decq %4; jnz "#ndim"008081b;"\
+    #ndim"008082:\n\t"\
+    SAVE_m8(ndim)
+
+/* m = 4 *//* zmm0 for alpha, zmm1-2 for perm words, zmm4-7 for temporary use, zmm8-15 for accumulators */
+#define KERNEL_k1m4n1 "vbroadcastss (%1),%%xmm4; addq $4,%1; vfmadd231ps (%0),%%xmm4,%%xmm8; addq $16,%0;"
+#define KERNEL_k1m4n2 "vmovups (%0),%%xmm4; addq $16,%0;"\
+    "vbroadcastss (%1),%%xmm5; vfmadd231ps %%xmm5,%%xmm4,%%xmm8;"\
+    "vbroadcastss 4(%1),%%xmm5; vfmadd231ps %%xmm5,%%xmm4,%%xmm9; addq $8,%1;"
+#define unit_kernel_k1m4n4(c1,...) "vbroadcastf32x4 ("#__VA_ARGS__"),%%zmm7; vfmadd231ps %%zmm7,%%zmm4,"#c1";"
+#define KERNEL_h_k1m4n4 "vbroadcastf32x4 (%0),%%zmm4; vpermilps %%zmm2,%%zmm4,%%zmm4; addq $16,%0;" unit_kernel_k1m4n4(%%zmm8,%1)
+#define KERNEL_k1m4n4 KERNEL_h_k1m4n4 "addq $16,%1;"
+#define KERNEL_h_k1m4n8 KERNEL_h_k1m4n4 unit_kernel_k1m4n4(%%zmm9,%1,%%r12,1)
+#define KERNEL_k1m4n8 KERNEL_h_k1m4n8 "addq $16,%1;"
+#define KERNEL_k1m4n12 KERNEL_h_k1m4n8 unit_kernel_k1m4n4(%%zmm10,%1,%%r12,2) "addq $16,%1;"
+#define KERNEL_h_k1m4n16 KERNEL_k1m4n12 unit_kernel_k1m4n4(%%zmm11,%%r15)
+#define KERNEL_k1m4n16 KERNEL_h_k1m4n16 "addq $16,%%r15;"
+#define KERNEL_h_k1m4n20 KERNEL_h_k1m4n16 unit_kernel_k1m4n4(%%zmm12,%%r15,%%r12,1)
+#define KERNEL_k1m4n20 KERNEL_h_k1m4n20 "addq $16,%%r15;"
+#define KERNEL_h_k1m4n24 KERNEL_h_k1m4n20 unit_kernel_k1m4n4(%%zmm13,%%r15,%%r12,2)
+#define KERNEL_k1m4n24 KERNEL_h_k1m4n24 "addq $16,%%r15;"
+#define INIT_m4n1 "vpxor %%xmm8,%%xmm8,%%xmm8;"
+#define INIT_m4n2 "vpxor %%xmm8,%%xmm8,%%xmm8; vpxor %%xmm9,%%xmm9,%%xmm9;"
+#define INIT_m4n4 "vpxorq %%zmm8,%%zmm8,%%zmm8;"
+#define INIT_m4n8 INIT_m4n4 "vpxorq %%zmm9,%%zmm9,%%zmm9;"
+#define INIT_m4n12 INIT_m4n8 "vpxorq %%zmm10,%%zmm10,%%zmm10;"
+#define INIT_m4n16 INIT_m4n12 "vpxorq %%zmm11,%%zmm11,%%zmm11;"
+#define INIT_m4n20 INIT_m4n16 "vpxorq %%zmm12,%%zmm12,%%zmm12;"
+#define INIT_m4n24 INIT_m4n20 "vpxorq %%zmm13,%%zmm13,%%zmm13;"
+#define SAVE_h_m4n1 "vfmadd213ps (%2),%%xmm0,%%xmm8; vmovups %%xmm8,(%2);"
+#define SAVE_h_m4n2 "vfmadd213ps (%2),%%xmm0,%%xmm8; vmovups %%xmm8,(%2); vfmadd213ps (%2,%3,1),%%xmm0,%%xmm9; vmovups %%xmm9,(%2,%3,1);"
+#define unit_save_m4n4(c1_no)\
+    "vpermps %%zmm"#c1_no",%%zmm1,%%zmm"#c1_no"; vextractf64x4 $1,%%zmm"#c1_no",%%ymm5;"\
+    "vmovups (%5),%%xmm4; vinsertf128 $1,(%5,%3,1),%%ymm4,%%ymm4; vfmadd231ps %%ymm0,%%ymm"#c1_no",%%ymm4;"\
+    "vmovups %%xmm4,(%5); vextractf128 $1,%%ymm4,(%5,%3,1); leaq (%5,%3,2),%5;"\
+    "vmovups (%5),%%xmm4; vinsertf128 $1,(%5,%3,1),%%ymm4,%%ymm4; vfmadd231ps %%ymm0,%%ymm5,%%ymm4;"\
+    "vmovups %%xmm4,(%5); vextractf128 $1,%%ymm4,(%5,%3,1); leaq (%5,%3,2),%5;"
+#define SAVE_h_m4n4 "movq %2,%5;" unit_save_m4n4(8)
+#define SAVE_h_m4n8 SAVE_h_m4n4 unit_save_m4n4(9)
+#define SAVE_h_m4n12 SAVE_h_m4n8 unit_save_m4n4(10)
+#define SAVE_h_m4n16 SAVE_h_m4n12 unit_save_m4n4(11)
+#define SAVE_h_m4n20 SAVE_h_m4n16 unit_save_m4n4(12)
+#define SAVE_h_m4n24 SAVE_h_m4n20 unit_save_m4n4(13)
+#define SAVE_m4(ndim) SAVE_h_m4n##ndim "addq $16,%2;"
+#define COMPUTE_m4(ndim) \
+    INIT_m4n##ndim\
+    "movq %%r13,%4; movq %%r14,%1; leaq (%1,%%r12,2),%%r15; addq %%r12,%%r15;"\
+    "testq %4,%4; jz "#ndim"004042f;"\
+    #ndim"004041:\n\t"\
+    KERNEL_k1m4n##ndim "decq %4; jnz "#ndim"004041b;"\
+    #ndim"004042:\n\t"\
+    SAVE_m4(ndim)
+
+/* m = 2 *//* xmm0 for alpha, xmm1-xmm3 for temporary use, xmm4-xmm15 for accumulators */
+#define INIT_m2n1 "vpxor %%xmm4,%%xmm4,%%xmm4;"
+#define KERNEL_k1m2n1 \
+    "vmovsd (%0),%%xmm1; addq $8,%0;"\
+    "vbroadcastss (%1),%%xmm2; vfmadd231ps %%xmm1,%%xmm2,%%xmm4;"\
+    "addq $4,%1;"
+#define SAVE_h_m2n1 "vmovsd (%2),%%xmm1; vfmadd213ps %%xmm1,%%xmm0,%%xmm4; vmovsd %%xmm4,(%2);"
+#define INIT_m2n2 INIT_m2n1 "vpxor %%xmm5,%%xmm5,%%xmm5;"
+#define KERNEL_k1m2n2 \
+    "vmovsd (%0),%%xmm1; addq $8,%0;"\
+    "vbroadcastss  (%1),%%xmm2; vfmadd231ps %%xmm1,%%xmm2,%%xmm4;"\
+    "vbroadcastss 4(%1),%%xmm3; vfmadd231ps %%xmm1,%%xmm3,%%xmm5;"\
+    "addq $8,%1;"
+#define SAVE_h_m2n2 SAVE_h_m2n1 "vmovsd (%2,%3,1),%%xmm1; vfmadd213ps %%xmm1,%%xmm0,%%xmm5; vmovsd %%xmm5,(%2,%3,1);"
+#define INIT_m2n4  INIT_m2n2
+#define INIT_m2n8  INIT_m2n4 "vpxor %%xmm6,%%xmm6,%%xmm6; vpxor %%xmm7,%%xmm7,%%xmm7;"
+#define INIT_m2n12 INIT_m2n8 "vpxor %%xmm8,%%xmm8,%%xmm8; vpxor %%xmm9,%%xmm9,%%xmm9;"
+#define INIT_m2n16 INIT_m2n12 "vpxor %%xmm10,%%xmm10,%%xmm10; vpxor %%xmm11,%%xmm11,%%xmm11;"
+#define INIT_m2n20 INIT_m2n16 "vpxor %%xmm12,%%xmm12,%%xmm12; vpxor %%xmm13,%%xmm13,%%xmm13;"
+#define INIT_m2n24 INIT_m2n20 "vpxor %%xmm14,%%xmm14,%%xmm14; vpxor %%xmm15,%%xmm15,%%xmm15;"
+#define KERNEL_h_k1m2n4 \
+    "vbroadcastss (%0),%%xmm1; vbroadcastss 4(%0),%%xmm2; addq $8,%0;"\
+    "vmovups (%1),%%xmm3; vfmadd231ps %%xmm1,%%xmm3,%%xmm4; vfmadd231ps %%xmm2,%%xmm3,%%xmm5;"
+#define KERNEL_k1m2n4 KERNEL_h_k1m2n4 "addq $16,%1;"
+#define KERNEL_h_k1m2n8 KERNEL_h_k1m2n4 "vmovups (%1,%%r12,1),%%xmm3; vfmadd231ps %%xmm1,%%xmm3,%%xmm6; vfmadd231ps %%xmm2,%%xmm3,%%xmm7;"
+#define KERNEL_k1m2n8 KERNEL_h_k1m2n8 "addq $16,%1;"
+#define KERNEL_k1m2n12 KERNEL_h_k1m2n8 \
+    "vmovups (%1,%%r12,2),%%xmm3; vfmadd231ps %%xmm1,%%xmm3,%%xmm8; vfmadd231ps %%xmm2,%%xmm3,%%xmm9; addq $16,%1;"
+#define KERNEL_h_k1m2n16 KERNEL_k1m2n12 "vmovups (%%r15),%%xmm3; vfmadd231ps %%xmm1,%%xmm3,%%xmm10; vfmadd231ps %%xmm2,%%xmm3,%%xmm11;"
+#define KERNEL_k1m2n16 KERNEL_h_k1m2n16 "addq $16,%%r15;"
+#define KERNEL_h_k1m2n20 KERNEL_h_k1m2n16 "vmovups (%%r15,%%r12,1),%%xmm3; vfmadd231ps %%xmm1,%%xmm3,%%xmm12; vfmadd231ps %%xmm2,%%xmm3,%%xmm13;"
+#define KERNEL_k1m2n20 KERNEL_h_k1m2n20 "addq $16,%%r15;"
+#define KERNEL_h_k1m2n24 KERNEL_h_k1m2n20 "vmovups (%%r15,%%r12,2),%%xmm3; vfmadd231ps %%xmm1,%%xmm3,%%xmm14; vfmadd231ps %%xmm2,%%xmm3,%%xmm15;"
+#define KERNEL_k1m2n24 KERNEL_h_k1m2n24 "addq $16,%%r15;"
+#define unit_save_m2n4(c1,c2) \
+    "vunpcklps "#c2","#c1",%%xmm1; vunpckhps "#c2","#c1",%%xmm2;"\
+    "vmovsd (%5),%%xmm3; vmovhpd (%5,%3,1),%%xmm3,%%xmm3; vfmadd213ps %%xmm3,%%xmm0,%%xmm1; vmovsd %%xmm1,(%5); vmovhpd %%xmm1,(%5,%3,1);"\
+    "leaq (%5,%3,2),%5;"\
+    "vmovsd (%5),%%xmm3; vmovhpd (%5,%3,1),%%xmm3,%%xmm3; vfmadd213ps %%xmm3,%%xmm0,%%xmm2; vmovsd %%xmm2,(%5); vmovhpd %%xmm2,(%5,%3,1);"\
+    "leaq (%5,%3,2),%5;"
+#define SAVE_h_m2n4  "movq %2,%5;" unit_save_m2n4(%%xmm4,%%xmm5)
+#define SAVE_h_m2n8  SAVE_h_m2n4   unit_save_m2n4(%%xmm6,%%xmm7)
+#define SAVE_h_m2n12 SAVE_h_m2n8   unit_save_m2n4(%%xmm8,%%xmm9)
+#define SAVE_h_m2n16 SAVE_h_m2n12  unit_save_m2n4(%%xmm10,%%xmm11)
+#define SAVE_h_m2n20 SAVE_h_m2n16  unit_save_m2n4(%%xmm12,%%xmm13)
+#define SAVE_h_m2n24 SAVE_h_m2n20  unit_save_m2n4(%%xmm14,%%xmm15)
+#define SAVE_m2(ndim) SAVE_h_m2n##ndim "addq $8,%2;"
+#define COMPUTE_m2(ndim) \
+    INIT_m2n##ndim\
+    "movq %%r13,%4; movq %%r14,%1; leaq (%1,%%r12,2),%%r15; addq %%r12,%%r15;"\
+    "testq %4,%4; jz "#ndim"002022f;"\
+    #ndim"002021:\n\t"\
+    KERNEL_k1m2n##ndim "decq %4; jnz "#ndim"002021b;"\
+    #ndim"002022:\n\t"\
+    SAVE_m2(ndim)
+
+/* m = 1 *//* xmm0 for alpha, xmm1-xmm3 and xmm10 for temporary use, xmm4-xmm9 for accumulators */
+#define INIT_m1n1 "vpxor %%xmm4,%%xmm4,%%xmm4;"
+#define KERNEL_k1m1n1 \
+    "vmovss (%1),%%xmm3; addq $4,%1;"\
+    "vmovss (%0),%%xmm1; vfmadd231ss %%xmm3,%%xmm1,%%xmm4;"\
+    "addq $4,%0;"
+#define SAVE_h_m1n1 "vfmadd213ss (%2),%%xmm0,%%xmm4; vmovss %%xmm4,(%2);"
+#define INIT_m1n2 INIT_m1n1
+#define KERNEL_k1m1n2 \
+    "vmovsd (%1),%%xmm3; addq $8,%1;"\
+    "vbroadcastss  (%0),%%xmm1; vfmadd231ps %%xmm3,%%xmm1,%%xmm4;"\
+    "addq $4,%0;"
+#define SAVE_h_m1n2 \
+    "vmovss (%2),%%xmm3; vinsertps $16,(%2,%3,1),%%xmm3,%%xmm3; vfmadd213ps %%xmm3,%%xmm0,%%xmm4;"\
+    "vmovss %%xmm4,(%2); vextractps $1,%%xmm4,(%2,%3,1);"
+#define INIT_m1n4  INIT_m1n2
+#define INIT_m1n8  INIT_m1n4 "vpxor %%xmm5,%%xmm5,%%xmm5;"
+#define INIT_m1n12 INIT_m1n8 "vpxor %%xmm6,%%xmm6,%%xmm6;"
+#define INIT_m1n16 INIT_m1n12 "vpxor %%xmm7,%%xmm7,%%xmm7;"
+#define INIT_m1n20 INIT_m1n16 "vpxor %%xmm8,%%xmm8,%%xmm8;"
+#define INIT_m1n24 INIT_m1n20 "vpxor %%xmm9,%%xmm9,%%xmm9;"
+#define KERNEL_h_k1m1n4 \
+    "vbroadcastss (%0),%%xmm1; addq $4,%0; vfmadd231ps (%1),%%xmm1,%%xmm4;"
+#define KERNEL_k1m1n4 KERNEL_h_k1m1n4 "addq $16,%1;"
+#define KERNEL_h_k1m1n8 KERNEL_h_k1m1n4 "vfmadd231ps (%1,%%r12,1),%%xmm1,%%xmm5;"
+#define KERNEL_k1m1n8 KERNEL_h_k1m1n8 "addq $16,%1;"
+#define KERNEL_k1m1n12 KERNEL_h_k1m1n8 "vfmadd231ps (%1,%%r12,2),%%xmm1,%%xmm6; addq $16,%1;"
+#define KERNEL_h_k1m1n16 KERNEL_k1m1n12 "vfmadd231ps (%%r15),%%xmm1,%%xmm7;"
+#define KERNEL_k1m1n16 KERNEL_h_k1m1n16 "addq $16,%%r15;"
+#define KERNEL_h_k1m1n20 KERNEL_h_k1m1n16 "vfmadd231ps (%%r15,%%r12,1),%%xmm1,%%xmm8;"
+#define KERNEL_k1m1n20 KERNEL_h_k1m1n20 "addq $16,%%r15;"
+#define KERNEL_h_k1m1n24 KERNEL_h_k1m1n20 "vfmadd231ps (%%r15,%%r12,2),%%xmm1,%%xmm9;"
+#define KERNEL_k1m1n24 KERNEL_h_k1m1n24 "addq $16,%%r15;"
+#define unit_save_m1n4(c1) \
+    "vpxor %%xmm10,%%xmm10,%%xmm10; vmovsd "#c1",%%xmm10,%%xmm2; vmovhlps "#c1",%%xmm10,%%xmm1;"\
+    "vmovss (%5),%%xmm3; vinsertps $16,(%5,%3,1),%%xmm3,%%xmm3; vfmadd213ps %%xmm3,%%xmm0,%%xmm2;"\
+    "vmovss %%xmm2,(%5); vextractps $1,%%xmm2,(%5,%3,1); leaq (%5,%3,2),%5;"\
+    "vmovss (%5),%%xmm3; vinsertps $16,(%5,%3,1),%%xmm3,%%xmm3; vfmadd213ps %%xmm3,%%xmm0,%%xmm1;"\
+    "vmovss %%xmm1,(%5); vextractps $1,%%xmm1,(%5,%3,1); leaq (%5,%3,2),%5;"
+#define SAVE_h_m1n4 "movq %2,%5;" unit_save_m1n4(%%xmm4)
+#define SAVE_h_m1n8  SAVE_h_m1n4  unit_save_m1n4(%%xmm5)
+#define SAVE_h_m1n12 SAVE_h_m1n8  unit_save_m1n4(%%xmm6)
+#define SAVE_h_m1n16 SAVE_h_m1n12 unit_save_m1n4(%%xmm7)
+#define SAVE_h_m1n20 SAVE_h_m1n16 unit_save_m1n4(%%xmm8)
+#define SAVE_h_m1n24 SAVE_h_m1n20 unit_save_m1n4(%%xmm9)
+#define SAVE_m1(ndim) SAVE_h_m1n##ndim "addq $4,%2;"
+#define COMPUTE_m1(ndim) \
+    INIT_m1n##ndim\
+    "movq %%r13,%4; movq %%r14,%1; leaq (%1,%%r12,2),%%r15; addq %%r12,%%r15;"\
+    "testq %4,%4; jz "#ndim"001012f;"\
+    #ndim"001011:\n\t"\
+    KERNEL_k1m1n##ndim "decq %4; jnz "#ndim"001011b;"\
+    #ndim"001012:\n\t"\
+    SAVE_m1(ndim)
+
+/* %0 = "+r"(a_pointer), %1 = "+r"(b_pointer), %2 = "+r"(c_pointer), %3 = "+r"(ldc_in_bytes), %4 = "+r"(K), %5 = "+r"(ctemp) */
+/* %6 = "+r"(next_b), %7 = "m"(ALPHA), %8 = "m"(M) */
+/* r11 = m_counter, r12 = k << 4(const), r13 = k(const), r14 = b_head_pos(const), r15 = %1 + 3r12 */
+
+#define COMPUTE(ndim) {\
+    next_b = b_pointer + ndim * K;\
+    __asm__ __volatile__(\
+    "vbroadcastss %7,%%zmm0; vmovups %9,%%zmm1; vmovups %10,%%zmm2;"\
+    "movq %4,%%r13; movq %4,%%r12; salq $4,%%r12; movq %1,%%r14; movq %8,%%r11;"\
+    "cmpq $16,%%r11;jb 33101"#ndim"f;"\
+    "33109"#ndim":\n\t"\
+    COMPUTE_m16(ndim)\
+    "subq $16,%%r11;cmpq $16,%%r11;jnb 33109"#ndim"b;"\
+    "33101"#ndim":\n\t"\
+    "cmpq $8,%%r11;jb 33102"#ndim"f;"\
+    COMPUTE_m8(ndim)\
+    "subq $8,%%r11;"\
+    "33102"#ndim":\n\t"\
+    "cmpq $4,%%r11;jb 33103"#ndim"f;"\
+    COMPUTE_m4(ndim)\
+    "subq $4,%%r11;"\
+    "33103"#ndim":\n\t"\
+    "cmpq $2,%%r11;jb 33104"#ndim"f;"\
+    COMPUTE_m2(ndim)\
+    "subq $2,%%r11;"\
+    "33104"#ndim":\n\t"\
+    "testq %%r11,%%r11;jz 33105"#ndim"f;"\
+    COMPUTE_m1(ndim)\
+    "33105"#ndim":\n\t"\
+    "movq %%r13,%4; movq %%r14,%1; vzeroupper;"\
+    :"+r"(a_pointer),"+r"(b_pointer),"+r"(c_pointer),"+r"(ldc_in_bytes),"+r"(K),"+r"(ctemp),"+r"(next_b):"m"(ALPHA),"m"(M),"m"(perm[0]),"m"(permil[0])\
+    :"r10","r11","r12","r13","r14","r15","zmm0","zmm1","zmm2","zmm3","zmm4","zmm5","zmm6","zmm7","zmm8","zmm9","zmm10","zmm11","zmm12","zmm13","zmm14",\
+    "zmm15","zmm16","zmm17","zmm18","zmm19","zmm20","zmm21","zmm22","zmm23","zmm24","zmm25","zmm26","zmm27","zmm28","zmm29","zmm30","zmm31",\
+    "cc","memory");\
+    a_pointer -= M * K; b_pointer += ndim * K; c_pointer += LDC * ndim - M;\
+}
+
+#define COMPUTE_n24 {\
+    next_b = b_pointer + 24 * K;\
+    __asm__ __volatile__(\
+    "vbroadcastss %8,%%zmm0; vmovups %10,%%zmm1; vmovups %11,%%zmm2;"\
+    "movq %4,%%r13; movq %4,%%r12; salq $4,%%r12; movq %1,%%r14; movq %9,%%r11;"\
+    "cmpq $32,%%r11;jb 3310024f;"\
+    COMPUTE_m16n24_LINIT "subq $16,%%r11; cmpq $32,%%r11;jb 3310724f;"\
+    "3310924:\n\t"\
+    COMPUTE_m16n24_RSAVE "subq $16,%%r11; cmpq $32,%%r11;jb 3310824f;"\
+    COMPUTE_m16n24_LSAVE "subq $16,%%r11; cmpq $32,%%r11;jnb 3310924b;"\
+    "3310724:\n\t"\
+    COMPUTE_m16n24_RTAIL "subq $16,%%r11; jmp 3310124f;"\
+    "3310824:\n\t"\
+    COMPUTE_m16n24_LTAIL "subq $16,%%r11; jmp 3310124f;"\
+    "3310024:\n\t"\
+    "cmpq $16,%%r11;jb 3310124f;"\
+    COMPUTE_m16(24)\
+    "subq $16,%%r11;"\
+    "3310124:\n\t"\
+    "cmpq $8,%%r11;jb 3310224f;"\
+    COMPUTE_m8(24)\
+    "subq $8,%%r11;"\
+    "3310224:\n\t"\
+    "cmpq $4,%%r11;jb 3310324f;"\
+    COMPUTE_m4(24)\
+    "subq $4,%%r11;"\
+    "3310324:\n\t"\
+    "cmpq $2,%%r11;jb 3310424f;"\
+    COMPUTE_m2(24)\
+    "subq $2,%%r11;"\
+    "3310424:\n\t"\
+    "testq %%r11,%%r11;jz 3310524f;"\
+    COMPUTE_m1(24)\
+    "3310524:\n\t"\
+    "movq %%r13,%4; movq %%r14,%1; vzeroupper;"\
+    :"+r"(a_pointer),"+r"(b_pointer),"+r"(c_pointer),"+r"(ldc_in_bytes),"+r"(K),"+r"(ctemp),"+r"(next_b),"+r"(wscr):"m"(ALPHA),"m"(M),"m"(perm[0]),"m"(permil[0])\
+    :"r10","r11","r12","r13","r14","r15","zmm0","zmm1","zmm2","zmm3","zmm4","zmm5","zmm6","zmm7","zmm8","zmm9","zmm10","zmm11","zmm12","zmm13","zmm14",\
+    "zmm15","zmm16","zmm17","zmm18","zmm19","zmm20","zmm21","zmm22","zmm23","zmm24","zmm25","zmm26","zmm27","zmm28","zmm29","zmm30","zmm31",\
+    "cc","memory");\
+    a_pointer -= M * K; b_pointer += 24 * K; c_pointer += LDC * 24 - M;\
+}
+
+int __attribute__ ((noinline))
+CNAME(BLASLONG m, BLASLONG n, BLASLONG k, float alpha, float * __restrict__ A, float * __restrict__ B, float * __restrict__ C, BLASLONG LDC)
+{
+    if(m==0||n==0||k==0||alpha==(float)0.0) return 0;
+    float scr[192]; float *wscr = scr;
+    int64_t ldc_in_bytes = (int64_t)LDC * sizeof(float);float ALPHA = alpha;
+    int64_t M = (int64_t)m, K = (int64_t)k;
+    int32_t perm[16] = {0,4,8,12,1,5,9,13,2,6,10,14,3,7,11,15};
+    int32_t permil[16] = {0,0,0,0,1,1,1,1,2,2,2,2,3,3,3,3};
+    BLASLONG n_count = n;
+    float *a_pointer = A,*b_pointer = B,*c_pointer = C,*ctemp = C,*next_b = B;
+    for(;n_count>23;n_count-=24) COMPUTE_n24
+    for(;n_count>19;n_count-=20) COMPUTE(20)
+    for(;n_count>15;n_count-=16) COMPUTE(16)
+    for(;n_count>11;n_count-=12) COMPUTE(12)
+    for(;n_count>7;n_count-=8) COMPUTE(8)
+    for(;n_count>3;n_count-=4) COMPUTE(4)
+    for(;n_count>1;n_count-=2) COMPUTE(2)
+    if(n_count>0) COMPUTE(1)
+    return 0;
+}
+#include <immintrin.h>
+#include "sgemm_direct_skylakex.c"
diff --git a/kernel/x86_64/strsm_kernel_8x4_haswell_LN.c b/kernel/x86_64/strsm_kernel_8x4_haswell_LN.c
new file mode 100644
index 000000000..4131debb1
--- /dev/null
+++ b/kernel/x86_64/strsm_kernel_8x4_haswell_LN.c
@@ -0,0 +1,240 @@
+#include "common.h"
+#include <stdint.h>
+#include "strsm_kernel_8x4_haswell_L_common.h"
+
+#define SOLVE_LN_m1n4 \
+  "subq $4,%2; movq %2,%3;" GEMM_SUM_REORDER_1x4(4)\
+  SOLVE_m1n4(-4,4) SAVE_b_m1n4(-16,4)\
+  "movq %2,%3;" save_c_m1n4(4)
+
+#define SOLVE_LN_m1n8 \
+  "subq $4,%2; movq %2,%3;" GEMM_SUM_REORDER_1x4(4) GEMM_SUM_REORDER_1x4(5)\
+  SOLVE_m1n8(-4,4,5) SAVE_b_m1n8(-16,4,5)\
+  "movq %2,%3;" save_c_m1n4(4) save_c_m1n4(5)
+
+#define SOLVE_LN_m1n12 \
+  "subq $4,%2; movq %2,%3;" GEMM_SUM_REORDER_1x4(4) GEMM_SUM_REORDER_1x4(5) GEMM_SUM_REORDER_1x4(6)\
+  SOLVE_m1n12(-4,4,5,6) SAVE_b_m1n12(-16,4,5,6)\
+  "movq %2,%3;" save_c_m1n4(4) save_c_m1n4(5) save_c_m1n4(6)
+
+#define SOLVE_LN_m2n4 \
+  "subq $8,%2; movq %2,%3;" GEMM_SUM_REORDER_2x4(4,5,4)\
+  SOLVE_loup_m2n4(-8,4)\
+  SOLVE_up_m2n4(-16,4) SAVE_b_m2n4(-32,4)\
+  "movq %2,%3;" save_c_m2n4(4)
+
+#define SOLVE_LN_m2n8 \
+  "subq $8,%2; movq %2,%3;" GEMM_SUM_REORDER_2x4(4,5,4) GEMM_SUM_REORDER_2x4(6,7,5)\
+  SOLVE_loup_m2n8(-8,4,5)\
+  SOLVE_up_m2n8(-16,4,5) SAVE_b_m2n8(-32,4,5)\
+  "movq %2,%3;" save_c_m2n4(4) save_c_m2n4(5)
+
+#define SOLVE_LN_m2n12 \
+  "subq $8,%2; movq %2,%3;" GEMM_SUM_REORDER_2x4(4,5,4) GEMM_SUM_REORDER_2x4(6,7,5) GEMM_SUM_REORDER_2x4(8,9,6)\
+  SOLVE_loup_m2n12(-8,4,5,6)\
+  SOLVE_up_m2n12(-16,4,5,6) SAVE_b_m2n12(-32,4,5,6)\
+  "movq %2,%3;" save_c_m2n4(4) save_c_m2n4(5) save_c_m2n4(6)
+
+#define SOLVE_LN_m4n4 \
+  "subq $16,%2; movq %2,%3;" GEMM_SUM_REORDER_4x4(4,5,6,7,4,5)\
+\
+  SOLVE_loup_m2n4(-8,5) SUBTRACT_m2n4(-16,4)\
+  SOLVE_up_m2n4(-24,5) SUBTRACT_m2n4(-32,4) SAVE_b_m2n4(-32,5)\
+\
+  SOLVE_loup_m2n4(-48,4)\
+  SOLVE_up_m2n4(-64,4) SAVE_b_m2n4(-64,4)\
+\
+  "movq %2,%3;" save_c_m4n4(4,5)
+
+#define SOLVE_LN_m4n8 \
+  "subq $16,%2; movq %2,%3;" GEMM_SUM_REORDER_4x4(4,5,6,7,4,5) GEMM_SUM_REORDER_4x4(8,9,10,11,6,7)\
+\
+  SOLVE_loup_m2n8(-8,5,7) SUBTRACT_m2n8(-16,4,6)\
+  SOLVE_up_m2n8(-24,5,7) SUBTRACT_m2n8(-32,4,6) SAVE_b_m2n8(-32,5,7)\
+\
+  SOLVE_loup_m2n8(-48,4,6)\
+  SOLVE_up_m2n8(-64,4,6) SAVE_b_m2n8(-64,4,6)\
+\
+  "movq %2,%3;" save_c_m4n4(4,5) save_c_m4n4(6,7)
+
+#define SOLVE_LN_m4n12 \
+  "subq $16,%2; movq %2,%3;" GEMM_SUM_REORDER_4x4(4,5,6,7,4,5) GEMM_SUM_REORDER_4x4(8,9,10,11,6,7) GEMM_SUM_REORDER_4x4(12,13,14,15,8,9)\
+\
+  SOLVE_loup_m2n12(-8,5,7,9) SUBTRACT_m2n12(-16,4,6,8)\
+  SOLVE_up_m2n12(-24,5,7,9) SUBTRACT_m2n12(-32,4,6,8) SAVE_b_m2n12(-32,5,7,9)\
+\
+  SOLVE_loup_m2n12(-48,4,6,8)\
+  SOLVE_up_m2n12(-64,4,6,8) SAVE_b_m2n12(-64,4,6,8)\
+\
+  "movq %2,%3;" save_c_m4n4(4,5) save_c_m4n4(6,7) save_c_m4n4(8,9)
+
+#define SOLVE_LN_m8n4 \
+  "subq $32,%2; movq %2,%3;" GEMM_SUM_REORDER_8x4(4,5,6,7,-32)\
+\
+  SOLVE_loup_m2n4(-8,7) SUBTRACT_m2n4(-16,6) SUBTRACT_m2n4(-24,5) SUBTRACT_m2n4(-32,4)\
+  SOLVE_up_m2n4(-40,7) SUBTRACT_m2n4(-48,6) SUBTRACT_m2n4(-56,5) SUBTRACT_m2n4(-64,4) SAVE_b_m2n4(-32,7)\
+\
+  SOLVE_loup_m2n4(-80,6) SUBTRACT_m2n4(-88,5) SUBTRACT_m2n4(-96,4)\
+  SOLVE_up_m2n4(-112,6) SUBTRACT_m2n4(-120,5) SUBTRACT_m2n4(-128,4) SAVE_b_m2n4(-64,6)\
+\
+  SOLVE_loup_m2n4(-152,5) SUBTRACT_m2n4(-160,4)\
+  SOLVE_up_m2n4(-184,5) SUBTRACT_m2n4(-192,4) SAVE_b_m2n4(-96,5)\
+\
+  SOLVE_loup_m2n4(-224,4)\
+  SOLVE_up_m2n4(-256,4) SAVE_b_m2n4(-128,4)\
+\
+  "movq %2,%3;" save_c_m8n4(4,5,6,7)
+
+#define SOLVE_LN_m8n8 \
+  "subq $32,%2; movq %2,%3;" GEMM_SUM_REORDER_8x4(4,5,6,7,-32) GEMM_SUM_REORDER_8x4(8,9,10,11,-32)\
+\
+  SOLVE_loup_m2n8(-8,7,11) SUBTRACT_m2n8(-16,6,10) SUBTRACT_m2n8(-24,5,9) SUBTRACT_m2n8(-32,4,8)\
+  SOLVE_up_m2n8(-40,7,11) SUBTRACT_m2n8(-48,6,10) SUBTRACT_m2n8(-56,5,9) SUBTRACT_m2n8(-64,4,8) SAVE_b_m2n8(-32,7,11)\
+\
+  SOLVE_loup_m2n8(-80,6,10) SUBTRACT_m2n8(-88,5,9) SUBTRACT_m2n8(-96,4,8)\
+  SOLVE_up_m2n8(-112,6,10) SUBTRACT_m2n8(-120,5,9) SUBTRACT_m2n8(-128,4,8) SAVE_b_m2n8(-64,6,10)\
+\
+  SOLVE_loup_m2n8(-152,5,9) SUBTRACT_m2n8(-160,4,8)\
+  SOLVE_up_m2n8(-184,5,9) SUBTRACT_m2n8(-192,4,8) SAVE_b_m2n8(-96,5,9)\
+\
+  SOLVE_loup_m2n8(-224,4,8)\
+  SOLVE_up_m2n8(-256,4,8) SAVE_b_m2n8(-128,4,8)\
+\
+  "movq %2,%3;" save_c_m8n4(4,5,6,7) save_c_m8n4(8,9,10,11)
+
+#define SOLVE_LN_m8n12 \
+  "subq $32,%2; movq %2,%3;" GEMM_SUM_REORDER_8x4(4,5,6,7,-32) GEMM_SUM_REORDER_8x4(8,9,10,11,-32) GEMM_SUM_REORDER_8x4(12,13,14,15,-32)\
+\
+  SOLVE_loup_m2n12(-8,7,11,15) SUBTRACT_m2n12(-16,6,10,14) SUBTRACT_m2n12(-24,5,9,13) SUBTRACT_m2n12(-32,4,8,12)\
+  SOLVE_up_m2n12(-40,7,11,15) SUBTRACT_m2n12(-48,6,10,14) SUBTRACT_m2n12(-56,5,9,13) SUBTRACT_m2n12(-64,4,8,12) SAVE_b_m2n12(-32,7,11,15)\
+\
+  SOLVE_loup_m2n12(-80,6,10,14) SUBTRACT_m2n12(-88,5,9,13) SUBTRACT_m2n12(-96,4,8,12)\
+  SOLVE_up_m2n12(-112,6,10,14) SUBTRACT_m2n12(-120,5,9,13) SUBTRACT_m2n12(-128,4,8,12) SAVE_b_m2n12(-64,6,10,14)\
+\
+  SOLVE_loup_m2n12(-152,5,9,13) SUBTRACT_m2n12(-160,4,8,12)\
+  SOLVE_up_m2n12(-184,5,9,13) SUBTRACT_m2n12(-192,4,8,12) SAVE_b_m2n12(-96,5,9,13)\
+\
+  SOLVE_loup_m2n12(-224,4,8,12)\
+  SOLVE_up_m2n12(-256,4,8,12) SAVE_b_m2n12(-128,4,8,12)\
+\
+  "movq %2,%3;" save_c_m8n4(4,5,6,7) save_c_m8n4(8,9,10,11) save_c_m8n4(12,13,14,15)
+
+/* r13 = k-kk, r14 = b_tail, r15 = a_tail */
+
+#define GEMM_LN_SIMPLE(mdim,ndim) \
+  "movq %%r15,%0; negq %%r12; leaq (%%r15,%%r12,"#mdim"),%%r15; negq %%r12;"\
+  "movq %%r13,%5; addq $"#mdim",%%r13; movq %%r14,%1;" INIT_m##mdim##n##ndim\
+  "testq %5,%5; jz 2"#mdim""#ndim"2f;"\
+  "2"#mdim""#ndim"1:\n\t"\
+  "subq $16,%1; subq $"#mdim"*4,%0;" GEMM_KERNEL_k1m##mdim##n##ndim "decq %5; jnz 2"#mdim""#ndim"1b;"\
+  "2"#mdim""#ndim"2:\n\t"
+#define GEMM_LN_m8n4 GEMM_LN_SIMPLE(8,4)
+#define GEMM_LN_m8n8 GEMM_LN_SIMPLE(8,8)
+#define GEMM_LN_m8n12 \
+  "movq %%r15,%0; negq %%r12; leaq (%%r15,%%r12,8),%%r15; negq %%r12; movq %%r13,%5; addq $8,%%r13; movq %%r14,%1;" INIT_m8n12\
+  "cmpq $8,%5; jb 28122f;"\
+  "28121:\n\t"\
+  "prefetcht0 -384(%0); subq $32,%0; subq $16,%1;" GEMM_KERNEL_k1m8n12\
+                       "subq $32,%0; subq $16,%1;" GEMM_KERNEL_k1m8n12\
+  "prefetcht0 -384(%0); subq $32,%0; subq $16,%1;" GEMM_KERNEL_k1m8n12\
+                       "subq $32,%0; subq $16,%1;" GEMM_KERNEL_k1m8n12\
+  "prefetcht0 -384(%0); subq $32,%0; subq $16,%1;" GEMM_KERNEL_k1m8n12\
+                       "subq $32,%0; subq $16,%1;" GEMM_KERNEL_k1m8n12\
+  "prefetcht0 -384(%0); subq $32,%0; subq $16,%1;" GEMM_KERNEL_k1m8n12\
+                       "subq $32,%0; subq $16,%1;" GEMM_KERNEL_k1m8n12\
+  "subq $8,%5; cmpq $8,%5; jnb 28121b;"\
+  "28122:\n\t"\
+  "testq %5,%5; jz 28124f;"\
+  "28123:\n\t"\
+  "subq $32,%0; subq $16,%1;" GEMM_KERNEL_k1m8n12 "decq %5; jnz 28123b;"\
+  "28124:\n\t"
+#define GEMM_LN_m4n4 GEMM_LN_SIMPLE(4,4)
+#define GEMM_LN_m4n8 GEMM_LN_SIMPLE(4,8)
+#define GEMM_LN_m4n12 GEMM_LN_SIMPLE(4,12)
+#define GEMM_LN_m2n4 GEMM_LN_SIMPLE(2,4)
+#define GEMM_LN_m2n8 GEMM_LN_SIMPLE(2,8)
+#define GEMM_LN_m2n12 GEMM_LN_SIMPLE(2,12)
+#define GEMM_LN_m1n4 GEMM_LN_SIMPLE(1,4)
+#define GEMM_LN_m1n8 GEMM_LN_SIMPLE(1,8)
+#define GEMM_LN_m1n12 GEMM_LN_SIMPLE(1,12)
+
+#define COMPUTE(ndim) {\
+  c_ptr += M;\
+  __asm__ __volatile__(\
+    "movq %0,%%r15; movq %7,%%r13; movq %6,%%r12; salq $2,%%r12; leaq (%1,%%r12,4),%%r14; movq %10,%%r11;"\
+    "testq $1,%%r11; jz "#ndim"772f;"\
+    #ndim"771:\n\t"\
+    GEMM_LN_m1n##ndim SOLVE_LN_m1n##ndim "subq $1,%%r11;"\
+    #ndim"772:\n\t"\
+    "testq $2,%%r11; jz "#ndim"773f;"\
+    GEMM_LN_m2n##ndim SOLVE_LN_m2n##ndim "subq $2,%%r11;"\
+    #ndim"773:\n\t"\
+    "testq $4,%%r11; jz "#ndim"774f;"\
+    GEMM_LN_m4n##ndim SOLVE_LN_m4n##ndim "subq $4,%%r11;"\
+    #ndim"774:\n\t"\
+    "testq %%r11,%%r11; jz "#ndim"776f;"\
+    #ndim"775:\n\t"\
+    GEMM_LN_m8n##ndim SOLVE_LN_m8n##ndim "subq $8,%%r11; jnz "#ndim"775b;"\
+    #ndim"776:\n\t"\
+    "movq %%r15,%0; movq %%r14,%1; vzeroupper;"\
+  :"+r"(a_ptr),"+r"(b_ptr),"+r"(c_ptr),"+r"(c_tmp),"+r"(ldc_bytes),"+r"(k_cnt):"m"(K),"m"(kmkkinp),"m"(one[0]),"m"(zero[0]),"m"(M)\
+  :"r11","r12","r13","r14","r15","cc","memory",\
+  "xmm0","xmm1","xmm2","xmm3","xmm4","xmm5","xmm6","xmm7","xmm8","xmm9","xmm10","xmm11","xmm12","xmm13","xmm14","xmm15");\
+  a_ptr += M * K; b_ptr += (ndim-4) * K; c_ptr += ldc * ndim;\
+}
+static void solve_LN(BLASLONG m, BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc) {
+  FLOAT a0, b0;
+  int i, j, k;
+  for (i=m-1;i>=0;i--) {
+    a0 = a[i*m+i]; //reciprocal of the original value
+    for (j=0;j<n;j++) {
+      b0 = c[j*ldc+i]*a0;
+      c[j*ldc+i] = b[i*n+j] = b0;
+      for (k=0;k<i;k++) c[j*ldc+k] -= b0*a[i*m+k];
+    }
+  }
+}
+static void COMPUTE_EDGE_1_nchunk(BLASLONG m, BLASLONG n, FLOAT *sa, FLOAT *sb, FLOAT *C, BLASLONG ldc, BLASLONG k, BLASLONG offset) {
+  BLASLONG m_count = m, kk = m+offset; FLOAT *a_ptr = sa+m*k, *c_ptr = C+m;
+  if(m_count&1){
+    a_ptr-=k; c_ptr--;
+    if(k-kk>0) GEMM_KERNEL_N(1,n,k-kk,-1.0,a_ptr+kk*1,sb+kk*n,c_ptr,ldc);
+    solve_LN(1,n,a_ptr+(kk-1)*1,sb+(kk-1)*n,c_ptr,ldc);
+    kk -= 1;
+    m_count--;
+  }
+  if(m_count&2){
+    a_ptr-=k*2; c_ptr-=2;
+    if(k-kk>0) GEMM_KERNEL_N(2,n,k-kk,-1.0,a_ptr+kk*2,sb+kk*n,c_ptr,ldc);
+    solve_LN(2,n,a_ptr+(kk-2)*2,sb+(kk-2)*n,c_ptr,ldc);
+    kk -= 2;
+    m_count-=2;
+  }
+  if(m_count&4){
+    a_ptr-=k*4; c_ptr-=4;
+    if(k-kk>0) GEMM_KERNEL_N(4,n,k-kk,-1.0,a_ptr+kk*4,sb+kk*n,c_ptr,ldc);
+    solve_LN(4,n,a_ptr+(kk-4)*4,sb+(kk-4)*n,c_ptr,ldc);
+    kk -= 4;
+    m_count-=4;
+  }
+  for(;m_count>7;m_count-=8){
+    a_ptr-=k*8; c_ptr-=8;
+    if(k-kk>0) GEMM_KERNEL_N(8,n,k-kk,-1.0,a_ptr+kk*8,sb+kk*n,c_ptr,ldc);
+    solve_LN(8,n,a_ptr+(kk-8)*8,sb+(kk-8)*n,c_ptr,ldc);
+    kk -= 8;
+  }
+}
+int CNAME(BLASLONG m, BLASLONG n, BLASLONG k, FLOAT dummy1, FLOAT *sa, FLOAT *sb, FLOAT *C, BLASLONG ldc, BLASLONG offset){
+  float *a_ptr = sa+m*k, *b_ptr = sb, *c_ptr = C, *c_tmp = C;
+  float one[8] = {1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0};
+  float zero[8] = {0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0};
+  uint64_t ldc_bytes = (uint64_t)ldc * sizeof(float), K = (uint64_t)k, M = (uint64_t)m, kmkkinp = (uint64_t)(k-m-offset), k_cnt = 0;
+  BLASLONG n_count = n;
+  for(;n_count>11;n_count-=12) COMPUTE(12)
+  for(;n_count>7;n_count-=8) COMPUTE(8)
+  for(;n_count>3;n_count-=4) COMPUTE(4)
+  for(;n_count>1;n_count-=2) { COMPUTE_EDGE_1_nchunk(m,2,sa,b_ptr,c_ptr,ldc,k,offset); b_ptr += 2*k; c_ptr += ldc*2;}
+  if(n_count>0) COMPUTE_EDGE_1_nchunk(m,1,sa,b_ptr,c_ptr,ldc,k,offset);
+  return 0;
+}
+
diff --git a/kernel/x86_64/strsm_kernel_8x4_haswell_LT.c b/kernel/x86_64/strsm_kernel_8x4_haswell_LT.c
new file mode 100644
index 000000000..3c7e9e83b
--- /dev/null
+++ b/kernel/x86_64/strsm_kernel_8x4_haswell_LT.c
@@ -0,0 +1,228 @@
+#include "common.h"
+#include <stdint.h>
+#include "strsm_kernel_8x4_haswell_L_common.h"
+
+#define SOLVE_LT_m1n4 \
+  "movq %2,%3;" GEMM_SUM_REORDER_1x4(4)\
+  SOLVE_m1n4(0,4) SAVE_b_m1n4(0,4)\
+  "movq %2,%3; addq $4,%2;" save_c_m1n4(4)
+
+#define SOLVE_LT_m1n8 \
+  "movq %2,%3;" GEMM_SUM_REORDER_1x4(4) GEMM_SUM_REORDER_1x4(5)\
+  SOLVE_m1n8(0,4,5) SAVE_b_m1n8(0,4,5)\
+  "movq %2,%3; addq $4,%2;" save_c_m1n4(4) save_c_m1n4(5)
+
+#define SOLVE_LT_m1n12 \
+  "movq %2,%3;" GEMM_SUM_REORDER_1x4(4) GEMM_SUM_REORDER_1x4(5) GEMM_SUM_REORDER_1x4(6)\
+  SOLVE_m1n12(0,4,5,6) SAVE_b_m1n12(0,4,5,6)\
+  "movq %2,%3; addq $4,%2;" save_c_m1n4(4) save_c_m1n4(5) save_c_m1n4(6)
+
+#define SOLVE_LT_m2n4 \
+  "movq %2,%3;" GEMM_SUM_REORDER_2x4(4,5,4)\
+  SOLVE_uplo_m2n4(0,4)\
+  SOLVE_lo_m2n4(8,4) SAVE_b_m2n4(0,4)\
+  "movq %2,%3; addq $8,%2;" save_c_m2n4(4)
+
+#define SOLVE_LT_m2n8 \
+  "movq %2,%3;" GEMM_SUM_REORDER_2x4(4,5,4) GEMM_SUM_REORDER_2x4(6,7,5)\
+  SOLVE_uplo_m2n8(0,4,5)\
+  SOLVE_lo_m2n8(8,4,5) SAVE_b_m2n8(0,4,5)\
+  "movq %2,%3; addq $8,%2;" save_c_m2n4(4) save_c_m2n4(5)
+
+#define SOLVE_LT_m2n12 \
+  "movq %2,%3;" GEMM_SUM_REORDER_2x4(4,5,4) GEMM_SUM_REORDER_2x4(6,7,5) GEMM_SUM_REORDER_2x4(8,9,6)\
+  SOLVE_uplo_m2n12(0,4,5,6)\
+  SOLVE_lo_m2n12(8,4,5,6) SAVE_b_m2n12(0,4,5,6)\
+  "movq %2,%3; addq $8,%2;" save_c_m2n4(4) save_c_m2n4(5) save_c_m2n4(6)
+
+#define SOLVE_LT_m4n4 \
+  "movq %2,%3;" GEMM_SUM_REORDER_4x4(4,5,6,7,4,5)\
+\
+  SOLVE_uplo_m2n4(0,4) SUBTRACT_m2n4(8,5)\
+  SOLVE_lo_m2n4(16,4) SUBTRACT_m2n4(24,5) SAVE_b_m2n4(0,4)\
+\
+  SOLVE_uplo_m2n4(40,5)\
+  SOLVE_lo_m2n4(56,5) SAVE_b_m2n4(32,5)\
+\
+  "movq %2,%3; addq $16,%2;" save_c_m4n4(4,5)
+
+#define SOLVE_LT_m4n8 \
+  "movq %2,%3;" GEMM_SUM_REORDER_4x4(4,5,6,7,4,5) GEMM_SUM_REORDER_4x4(8,9,10,11,6,7)\
+\
+  SOLVE_uplo_m2n8(0,4,6) SUBTRACT_m2n8(8,5,7)\
+  SOLVE_lo_m2n8(16,4,6) SUBTRACT_m2n8(24,5,7) SAVE_b_m2n8(0,4,6)\
+\
+  SOLVE_uplo_m2n8(40,5,7)\
+  SOLVE_lo_m2n8(56,5,7) SAVE_b_m2n8(32,5,7)\
+\
+  "movq %2,%3; addq $16,%2;" save_c_m4n4(4,5) save_c_m4n4(6,7)
+
+#define SOLVE_LT_m4n12 \
+  "movq %2,%3;" GEMM_SUM_REORDER_4x4(4,5,6,7,4,5) GEMM_SUM_REORDER_4x4(8,9,10,11,6,7) GEMM_SUM_REORDER_4x4(12,13,14,15,8,9)\
+\
+  SOLVE_uplo_m2n12(0,4,6,8) SUBTRACT_m2n12(8,5,7,9)\
+  SOLVE_lo_m2n12(16,4,6,8) SUBTRACT_m2n12(24,5,7,9) SAVE_b_m2n12(0,4,6,8)\
+\
+  SOLVE_uplo_m2n12(40,5,7,9)\
+  SOLVE_lo_m2n12(56,5,7,9) SAVE_b_m2n12(32,5,7,9)\
+\
+  "movq %2,%3; addq $16,%2;" save_c_m4n4(4,5) save_c_m4n4(6,7) save_c_m4n4(8,9)
+
+#define SOLVE_LT_m8n4 \
+  "movq %2,%3;" GEMM_SUM_REORDER_8x4(4,5,6,7,63)\
+\
+  SOLVE_uplo_m2n4(0,4) SUBTRACT_m2n4(8,5) SUBTRACT_m2n4(16,6) SUBTRACT_m2n4(24,7)\
+  SOLVE_lo_m2n4(32,4) SUBTRACT_m2n4(40,5) SUBTRACT_m2n4(48,6) SUBTRACT_m2n4(56,7) SAVE_b_m2n4(0,4)\
+\
+  SOLVE_uplo_m2n4(72,5) SUBTRACT_m2n4(80,6) SUBTRACT_m2n4(88,7)\
+  SOLVE_lo_m2n4(104,5) SUBTRACT_m2n4(112,6) SUBTRACT_m2n4(120,7) SAVE_b_m2n4(32,5)\
+\
+  SOLVE_uplo_m2n4(144,6) SUBTRACT_m2n4(152,7)\
+  SOLVE_lo_m2n4(176,6) SUBTRACT_m2n4(184,7) SAVE_b_m2n4(64,6)\
+\
+  SOLVE_uplo_m2n4(216,7)\
+  SOLVE_lo_m2n4(248,7) SAVE_b_m2n4(96,7)\
+\
+  "movq %2,%3; addq $32,%2;" save_c_m8n4(4,5,6,7)
+
+#define SOLVE_LT_m8n8 \
+  "movq %2,%3;" GEMM_SUM_REORDER_8x4(4,5,6,7,63) GEMM_SUM_REORDER_8x4(8,9,10,11,63)\
+\
+  SOLVE_uplo_m2n8(0,4,8) SUBTRACT_m2n8(8,5,9) SUBTRACT_m2n8(16,6,10) SUBTRACT_m2n8(24,7,11)\
+  SOLVE_lo_m2n8(32,4,8) SUBTRACT_m2n8(40,5,9) SUBTRACT_m2n8(48,6,10) SUBTRACT_m2n8(56,7,11) SAVE_b_m2n8(0,4,8)\
+\
+  SOLVE_uplo_m2n8(72,5,9) SUBTRACT_m2n8(80,6,10) SUBTRACT_m2n8(88,7,11)\
+  SOLVE_lo_m2n8(104,5,9) SUBTRACT_m2n8(112,6,10) SUBTRACT_m2n8(120,7,11) SAVE_b_m2n8(32,5,9)\
+\
+  SOLVE_uplo_m2n8(144,6,10) SUBTRACT_m2n8(152,7,11)\
+  SOLVE_lo_m2n8(176,6,10) SUBTRACT_m2n8(184,7,11) SAVE_b_m2n8(64,6,10)\
+\
+  SOLVE_uplo_m2n8(216,7,11)\
+  SOLVE_lo_m2n8(248,7,11) SAVE_b_m2n8(96,7,11)\
+\
+  "movq %2,%3; addq $32,%2;" save_c_m8n4(4,5,6,7) save_c_m8n4(8,9,10,11)
+
+#define SOLVE_LT_m8n12 \
+  "movq %2,%3;" GEMM_SUM_REORDER_8x4(4,5,6,7,63) GEMM_SUM_REORDER_8x4(8,9,10,11,63) GEMM_SUM_REORDER_8x4(12,13,14,15,63)\
+\
+  SOLVE_uplo_m2n12(0,4,8,12) SUBTRACT_m2n12(8,5,9,13) SUBTRACT_m2n12(16,6,10,14) SUBTRACT_m2n12(24,7,11,15)\
+  SOLVE_lo_m2n12(32,4,8,12) SUBTRACT_m2n12(40,5,9,13) SUBTRACT_m2n12(48,6,10,14) SUBTRACT_m2n12(56,7,11,15) SAVE_b_m2n12(0,4,8,12)\
+\
+  SOLVE_uplo_m2n12(72,5,9,13) SUBTRACT_m2n12(80,6,10,14) SUBTRACT_m2n12(88,7,11,15)\
+  SOLVE_lo_m2n12(104,5,9,13) SUBTRACT_m2n12(112,6,10,14) SUBTRACT_m2n12(120,7,11,15) SAVE_b_m2n12(32,5,9,13)\
+\
+  SOLVE_uplo_m2n12(144,6,10,14) SUBTRACT_m2n12(152,7,11,15)\
+  SOLVE_lo_m2n12(176,6,10,14) SUBTRACT_m2n12(184,7,11,15) SAVE_b_m2n12(64,6,10,14)\
+\
+  SOLVE_uplo_m2n12(216,7,11,15)\
+  SOLVE_lo_m2n12(248,7,11,15) SAVE_b_m2n12(96,7,11,15)\
+\
+  "movq %2,%3; addq $32,%2;" save_c_m8n4(4,5,6,7) save_c_m8n4(8,9,10,11) save_c_m8n4(12,13,14,15)
+
+#define GEMM_LT_SIMPLE(mdim,ndim) \
+  "movq %%r15,%0; leaq (%%r15,%%r12,"#mdim"),%%r15; movq %%r13,%5; addq $"#mdim",%%r13; movq %%r14,%1;" INIT_m##mdim##n##ndim\
+  "testq %5,%5; jz 1"#mdim""#ndim"2f;"\
+  "1"#mdim""#ndim"1:\n\t"\
+  GEMM_KERNEL_k1m##mdim##n##ndim "addq $16,%1; addq $"#mdim"*4,%0; decq %5; jnz 1"#mdim""#ndim"1b;"\
+  "1"#mdim""#ndim"2:\n\t"
+#define GEMM_LT_m8n4 GEMM_LT_SIMPLE(8,4)
+#define GEMM_LT_m8n8 GEMM_LT_SIMPLE(8,8)
+#define GEMM_LT_m8n12 \
+  "movq %%r15,%0; leaq (%%r15,%%r12,8),%%r15; movq %%r13,%5; addq $8,%%r13; movq %%r14,%1;" INIT_m8n12\
+  "cmpq $8,%5; jb 18122f;"\
+  "18121:\n\t"\
+  GEMM_KERNEL_k1m8n12 "prefetcht0 384(%0); addq $32,%0; addq $16,%1;"\
+  GEMM_KERNEL_k1m8n12 "addq $32,%0; addq $16,%1;"\
+  GEMM_KERNEL_k1m8n12 "prefetcht0 384(%0); addq $32,%0; addq $16,%1;"\
+  GEMM_KERNEL_k1m8n12 "addq $32,%0; addq $16,%1;"\
+  GEMM_KERNEL_k1m8n12 "prefetcht0 384(%0); addq $32,%0; addq $16,%1;"\
+  GEMM_KERNEL_k1m8n12 "addq $32,%0; addq $16,%1;"\
+  GEMM_KERNEL_k1m8n12 "prefetcht0 384(%0); addq $32,%0; addq $16,%1;"\
+  GEMM_KERNEL_k1m8n12 "addq $32,%0; addq $16,%1;"\
+  "subq $8,%5; cmpq $8,%5; jnb 18121b;"\
+  "18122:\n\t"\
+  "testq %5,%5; jz 18124f;"\
+  "18123:\n\t"\
+  GEMM_KERNEL_k1m8n12 "addq $32,%0; addq $16,%1; decq %5; jnz 18123b;"\
+  "18124:\n\t"
+#define GEMM_LT_m4n4 GEMM_LT_SIMPLE(4,4)
+#define GEMM_LT_m4n8 GEMM_LT_SIMPLE(4,8)
+#define GEMM_LT_m4n12 GEMM_LT_SIMPLE(4,12)
+#define GEMM_LT_m2n4 GEMM_LT_SIMPLE(2,4)
+#define GEMM_LT_m2n8 GEMM_LT_SIMPLE(2,8)
+#define GEMM_LT_m2n12 GEMM_LT_SIMPLE(2,12)
+#define GEMM_LT_m1n4 GEMM_LT_SIMPLE(1,4)
+#define GEMM_LT_m1n8 GEMM_LT_SIMPLE(1,8)
+#define GEMM_LT_m1n12 GEMM_LT_SIMPLE(1,12)
+
+#define COMPUTE(ndim) {\
+  __asm__ __volatile__(\
+    "movq %0,%%r15; movq %1,%%r14; movq %7,%%r13; movq %6,%%r12; salq $2,%%r12; movq %10,%%r11;"\
+    "cmpq $8,%%r11; jb "#ndim"772f;"\
+    #ndim"771:\n\t"\
+    GEMM_LT_m8n##ndim SOLVE_LT_m8n##ndim "subq $8,%%r11; cmpq $8,%%r11; jnb "#ndim"771b;"\
+    #ndim"772:\n\t"\
+    "testq $4,%%r11; jz "#ndim"773f;"\
+    GEMM_LT_m4n##ndim SOLVE_LT_m4n##ndim "subq $4,%%r11;"\
+    #ndim"773:\n\t"\
+    "testq $2,%%r11; jz "#ndim"774f;"\
+    GEMM_LT_m2n##ndim SOLVE_LT_m2n##ndim "subq $2,%%r11;"\
+    #ndim"774:\n\t"\
+    "testq $1,%%r11; jz "#ndim"775f;"\
+    GEMM_LT_m1n##ndim SOLVE_LT_m1n##ndim "subq $1,%%r11;"\
+    #ndim"775:\n\t"\
+    "movq %%r15,%0; movq %%r14,%1; vzeroupper;"\
+  :"+r"(a_ptr),"+r"(b_ptr),"+r"(c_ptr),"+r"(c_tmp),"+r"(ldc_bytes),"+r"(k_cnt):"m"(K),"m"(OFF),"m"(one[0]),"m"(zero[0]),"m"(M)\
+  :"r11","r12","r13","r14","r15","cc","memory",\
+  "xmm0","xmm1","xmm2","xmm3","xmm4","xmm5","xmm6","xmm7","xmm8","xmm9","xmm10","xmm11","xmm12","xmm13","xmm14","xmm15");\
+  a_ptr -= M * K; b_ptr += ndim * K; c_ptr += ldc * ndim - M;\
+}
+static void solve_LT(BLASLONG m, BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc) {
+  FLOAT a0, b0;
+  int i, j, k;
+  for (i=0;i<m;i++) {
+    a0 = a[i*m+i];
+    for (j=0;j<n;j++) {
+      b0 = c[j*ldc+i] * a0;
+      b[i*n+j] = c[j*ldc+i] = b0;
+      for (k=i+1;k<m;k++) c[j*ldc+k] -= b0 * a[i*m+k];
+    }
+  }
+}
+static void COMPUTE_EDGE_1_nchunk(BLASLONG m, BLASLONG n, FLOAT *sa, FLOAT *sb, FLOAT *C, BLASLONG ldc, BLASLONG k, BLASLONG offset) {
+  BLASLONG m_count = m, kk = offset; FLOAT *a_ptr = sa, *c_ptr = C;
+  for(;m_count>7;m_count-=8){
+    if(kk>0) GEMM_KERNEL_N(8,n,kk,-1.0,a_ptr,sb,c_ptr,ldc);
+    solve_LT(8,n,a_ptr+kk*8,sb+kk*n,c_ptr,ldc);
+    kk += 8; a_ptr += k * 8; c_ptr += 8;
+  }
+  for(;m_count>3;m_count-=4){
+    if(kk>0) GEMM_KERNEL_N(4,n,kk,-1.0,a_ptr,sb,c_ptr,ldc);
+    solve_LT(4,n,a_ptr+kk*4,sb+kk*n,c_ptr,ldc);
+    kk += 4; a_ptr += k * 4; c_ptr += 4;
+  }
+  for(;m_count>1;m_count-=2){
+    if(kk>0) GEMM_KERNEL_N(2,n,kk,-1.0,a_ptr,sb,c_ptr,ldc);
+    solve_LT(2,n,a_ptr+kk*2,sb+kk*n,c_ptr,ldc);
+    kk += 2; a_ptr += k * 2; c_ptr += 2;
+  }
+  if(m_count>0){
+    if(kk>0) GEMM_KERNEL_N(1,n,kk,-1.0,a_ptr,sb,c_ptr,ldc);
+    solve_LT(1,n,a_ptr+kk*1,sb+kk*n,c_ptr,ldc);
+    kk += 1; a_ptr += k * 1; c_ptr += 1;
+  }
+}
+int CNAME(BLASLONG m, BLASLONG n, BLASLONG k, FLOAT dummy1, FLOAT *sa, FLOAT *sb, FLOAT *C, BLASLONG ldc, BLASLONG offset){
+  float *a_ptr = sa, *b_ptr = sb, *c_ptr = C, *c_tmp = C;
+  float one[8] = {1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0};
+  float zero[8] = {0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0};
+  uint64_t ldc_bytes = (uint64_t)ldc * sizeof(float), K = (uint64_t)k, M = (uint64_t)m, OFF = (uint64_t)offset, k_cnt = 0;
+  BLASLONG n_count = n;
+  for(;n_count>11;n_count-=12) COMPUTE(12)
+  for(;n_count>7;n_count-=8) COMPUTE(8)
+  for(;n_count>3;n_count-=4) COMPUTE(4)
+  for(;n_count>1;n_count-=2) { COMPUTE_EDGE_1_nchunk(m,2,a_ptr,b_ptr,c_ptr,ldc,k,offset); b_ptr += 2*k; c_ptr += ldc*2;}
+  if(n_count>0) COMPUTE_EDGE_1_nchunk(m,1,a_ptr,b_ptr,c_ptr,ldc,k,offset);
+  return 0;
+}
+
diff --git a/kernel/x86_64/strsm_kernel_8x4_haswell_L_common.h b/kernel/x86_64/strsm_kernel_8x4_haswell_L_common.h
new file mode 100644
index 000000000..cfa56da97
--- /dev/null
+++ b/kernel/x86_64/strsm_kernel_8x4_haswell_L_common.h
@@ -0,0 +1,187 @@
+/* r11 = m_counter, r12 = size_of_k_elements, r13 = kk, r14 = b_head, r15 = a_head */
+/* register i/o: %0 = a_ptr, %1 = b_ptr, %2 = c_ptr, %3 = c_tmp, %4 = ldc, %5 = k_counter */
+/* memory input: %6 = K, %7 = offset, %8 = {1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0}, %9 = {0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0}, %10 = M */
+
+#define init_m8n4(c1,c2,c3,c4)\
+  "vpxor %%ymm"#c1",%%ymm"#c1",%%ymm"#c1"; vpxor %%ymm"#c2",%%ymm"#c2",%%ymm"#c2"; vpxor %%ymm"#c3",%%ymm"#c3",%%ymm"#c3"; vpxor %%ymm"#c4",%%ymm"#c4",%%ymm"#c4";"
+#define INIT_m8n4 init_m8n4(4,5,6,7)
+#define INIT_m8n8 INIT_m8n4 init_m8n4(8,9,10,11)
+#define INIT_m8n12 INIT_m8n8 init_m8n4(12,13,14,15)
+
+#define init_m4n4(c1,c2,c3,c4)\
+  "vpxor %%xmm"#c1",%%xmm"#c1",%%xmm"#c1"; vpxor %%xmm"#c2",%%xmm"#c2",%%xmm"#c2"; vpxor %%xmm"#c3",%%xmm"#c3",%%xmm"#c3"; vpxor %%xmm"#c4",%%xmm"#c4",%%xmm"#c4";"
+#define INIT_m4n4 init_m4n4(4,5,6,7)
+#define INIT_m4n8 INIT_m4n4 init_m4n4(8,9,10,11)
+#define INIT_m4n12 INIT_m4n8 init_m4n4(12,13,14,15)
+
+#define init_m2n4(c1,c2)\
+  "vpxor %%xmm"#c1",%%xmm"#c1",%%xmm"#c1"; vpxor %%xmm"#c2",%%xmm"#c2",%%xmm"#c2";"
+#define INIT_m2n4 init_m2n4(4,5)
+#define INIT_m2n8 INIT_m2n4 init_m2n4(6,7)
+#define INIT_m2n12 INIT_m2n8 init_m2n4(8,9)
+
+#define init_m1n4(c1) "vpxor %%xmm"#c1",%%xmm"#c1",%%xmm"#c1";"
+#define INIT_m1n4 init_m1n4(4)
+#define INIT_m1n8 INIT_m1n4 init_m1n4(5)
+#define INIT_m1n12 INIT_m1n8 init_m1n4(6)
+
+#define GEMM_KERNEL_k1m8n4 \
+  "vmovsldup (%0),%%ymm1; vmovshdup (%0),%%ymm2;"\
+  "vbroadcastsd (%1),%%ymm3; vfnmadd231ps %%ymm3,%%ymm1,%%ymm4; vfnmadd231ps %%ymm3,%%ymm2,%%ymm5;"\
+  "vbroadcastsd 8(%1),%%ymm3; vfnmadd231ps %%ymm3,%%ymm1,%%ymm6; vfnmadd231ps %%ymm3,%%ymm2,%%ymm7;"
+#define GEMM_KERNEL_k1m8n8 GEMM_KERNEL_k1m8n4\
+  "vbroadcastsd (%1,%%r12,4),%%ymm3; vfnmadd231ps %%ymm3,%%ymm1,%%ymm8; vfnmadd231ps %%ymm3,%%ymm2,%%ymm9;"\
+  "vbroadcastsd 8(%1,%%r12,4),%%ymm3; vfnmadd231ps %%ymm3,%%ymm1,%%ymm10; vfnmadd231ps %%ymm3,%%ymm2,%%ymm11;"
+#define GEMM_KERNEL_k1m8n12 GEMM_KERNEL_k1m8n8\
+  "vbroadcastsd (%1,%%r12,8),%%ymm3; vfnmadd231ps %%ymm3,%%ymm1,%%ymm12; vfnmadd231ps %%ymm3,%%ymm2,%%ymm13;"\
+  "vbroadcastsd 8(%1,%%r12,8),%%ymm3; vfnmadd231ps %%ymm3,%%ymm1,%%ymm14; vfnmadd231ps %%ymm3,%%ymm2,%%ymm15;"
+
+#define GEMM_KERNEL_k1m4n4 \
+  "vmovsldup (%0),%%xmm1; vmovshdup (%0),%%xmm2;"\
+  "vmovddup (%1),%%xmm3; vfnmadd231ps %%xmm3,%%xmm1,%%xmm4; vfnmadd231ps %%xmm3,%%xmm2,%%xmm5;"\
+  "vmovddup 8(%1),%%xmm3; vfnmadd231ps %%xmm3,%%xmm1,%%xmm6; vfnmadd231ps %%xmm3,%%xmm2,%%xmm7;"
+#define GEMM_KERNEL_k1m4n8 GEMM_KERNEL_k1m4n4\
+  "vmovddup (%1,%%r12,4),%%xmm3; vfnmadd231ps %%xmm3,%%xmm1,%%xmm8; vfnmadd231ps %%xmm3,%%xmm2,%%xmm9;"\
+  "vmovddup 8(%1,%%r12,4),%%xmm3; vfnmadd231ps %%xmm3,%%xmm1,%%xmm10; vfnmadd231ps %%xmm3,%%xmm2,%%xmm11;"
+#define GEMM_KERNEL_k1m4n12 GEMM_KERNEL_k1m4n8\
+  "vmovddup (%1,%%r12,8),%%xmm3; vfnmadd231ps %%xmm3,%%xmm1,%%xmm12; vfnmadd231ps %%xmm3,%%xmm2,%%xmm13;"\
+  "vmovddup 8(%1,%%r12,8),%%xmm3; vfnmadd231ps %%xmm3,%%xmm1,%%xmm14; vfnmadd231ps %%xmm3,%%xmm2,%%xmm15;"
+
+#define GEMM_KERNEL_k1m2n4 \
+  "vbroadcastss (%0),%%xmm1; vbroadcastss 4(%0),%%xmm2;"\
+  "vmovups (%1),%%xmm3; vfnmadd231ps %%xmm3,%%xmm1,%%xmm4; vfnmadd231ps %%xmm3,%%xmm2,%%xmm5;"
+#define GEMM_KERNEL_k1m2n8 GEMM_KERNEL_k1m2n4\
+  "vmovups (%1,%%r12,4),%%xmm3; vfnmadd231ps %%xmm3,%%xmm1,%%xmm6; vfnmadd231ps %%xmm3,%%xmm2,%%xmm7;"
+#define GEMM_KERNEL_k1m2n12 GEMM_KERNEL_k1m2n8\
+  "vmovups (%1,%%r12,8),%%xmm3; vfnmadd231ps %%xmm3,%%xmm1,%%xmm8; vfnmadd231ps %%xmm3,%%xmm2,%%xmm9;"
+
+#define GEMM_KERNEL_k1m1n4 "vbroadcastss (%0),%%xmm1; vfnmadd231ps (%1),%%xmm1,%%xmm4;"
+#define GEMM_KERNEL_k1m1n8 GEMM_KERNEL_k1m1n4 "vfnmadd231ps (%1,%%r12,4),%%xmm1,%%xmm5;"
+#define GEMM_KERNEL_k1m1n12 GEMM_KERNEL_k1m1n8 "vfnmadd231ps (%1,%%r12,8),%%xmm1,%%xmm6;"
+
+#define GEMM_SUM_REORDER_8x4(c1,c2,c3,c4,prefpos)\
+  "vunpcklps %%ymm"#c2",%%ymm"#c1",%%ymm0; vunpckhps %%ymm"#c2",%%ymm"#c1",%%ymm1;"\
+  "vunpcklps %%ymm"#c4",%%ymm"#c3",%%ymm2; vunpckhps %%ymm"#c4",%%ymm"#c3",%%ymm3;"\
+  "vmovups (%3),%%ymm"#c1"; vmovups (%3,%4,1),%%ymm"#c2"; prefetcht1 "#prefpos"(%3); prefetcht1 "#prefpos"(%3,%4,1);"\
+  "vunpcklpd %%ymm"#c2",%%ymm"#c1",%%ymm"#c3"; vunpckhpd %%ymm"#c2",%%ymm"#c1",%%ymm"#c4";"\
+  "vaddps %%ymm0,%%ymm"#c3",%%ymm0; vaddps %%ymm1,%%ymm"#c4",%%ymm1;"\
+  "leaq (%3,%4,2),%3;"\
+  "vmovups (%3),%%ymm"#c1"; vmovups (%3,%4,1),%%ymm"#c2"; prefetcht1 "#prefpos"(%3); prefetcht1 "#prefpos"(%3,%4,1);"\
+  "vunpcklpd %%ymm"#c2",%%ymm"#c1",%%ymm"#c3"; vunpckhpd %%ymm"#c2",%%ymm"#c1",%%ymm"#c4";"\
+  "vaddps %%ymm2,%%ymm"#c3",%%ymm2; vaddps %%ymm3,%%ymm"#c4",%%ymm3;"\
+  "leaq (%3,%4,2),%3;"\
+  "vperm2f128 $2,%%ymm0,%%ymm2,%%ymm"#c1"; vperm2f128 $2,%%ymm1,%%ymm3,%%ymm"#c2";"\
+  "vperm2f128 $19,%%ymm0,%%ymm2,%%ymm"#c3"; vperm2f128 $19,%%ymm1,%%ymm3,%%ymm"#c4";"
+
+#define GEMM_SUM_REORDER_4x4(c1,c2,c3,c4,co1,co2)\
+  "vunpcklps %%xmm"#c2",%%xmm"#c1",%%xmm0; vunpckhps %%xmm"#c2",%%xmm"#c1",%%xmm1;"\
+  "vunpcklps %%xmm"#c4",%%xmm"#c3",%%xmm2; vunpckhps %%xmm"#c4",%%xmm"#c3",%%xmm3;"\
+  "vmovups (%3),%%xmm"#c1"; vmovups (%3,%4,1),%%xmm"#c2";"\
+  "vunpcklpd %%xmm"#c2",%%xmm"#c1",%%xmm"#c3"; vunpckhpd %%xmm"#c2",%%xmm"#c1",%%xmm"#c4";"\
+  "vaddps %%xmm0,%%xmm"#c3",%%xmm0; vaddps %%xmm1,%%xmm"#c4",%%xmm1;"\
+  "leaq (%3,%4,2),%3;"\
+  "vmovups (%3),%%xmm"#c1"; vmovups (%3,%4,1),%%xmm"#c2";"\
+  "vunpcklpd %%xmm"#c2",%%xmm"#c1",%%xmm"#c3"; vunpckhpd %%xmm"#c2",%%xmm"#c1",%%xmm"#c4";"\
+  "vaddps %%xmm2,%%xmm"#c3",%%xmm2; vaddps %%xmm3,%%xmm"#c4",%%xmm3;"\
+  "leaq (%3,%4,2),%3;"\
+  "vperm2f128 $2,%%ymm0,%%ymm2,%%ymm"#co1"; vperm2f128 $2,%%ymm1,%%ymm3,%%ymm"#co2";"
+
+#define GEMM_SUM_REORDER_2x4(c1,c2,co1)\
+  "vunpcklps %%xmm"#c2",%%xmm"#c1",%%xmm0; vunpckhps %%xmm"#c2",%%xmm"#c1",%%xmm1;"\
+  "vmovsd (%3),%%xmm2; vmovhpd (%3,%4,1),%%xmm2,%%xmm2; vaddps %%xmm0,%%xmm2,%%xmm0; leaq (%3,%4,2),%3;"\
+  "vmovsd (%3),%%xmm2; vmovhpd (%3,%4,1),%%xmm2,%%xmm2; vaddps %%xmm1,%%xmm2,%%xmm1; leaq (%3,%4,2),%3;"\
+  "vperm2f128 $2,%%ymm0,%%ymm1,%%ymm"#co1";"
+
+#define GEMM_SUM_REORDER_1x4(c1)\
+  "vmovss (%3),%%xmm1; vinsertps $16,(%3,%4,1),%%xmm1,%%xmm1; leaq (%3,%4,2),%3;"\
+  "vinsertps $32,(%3),%%xmm1,%%xmm1; vinsertps $48,(%3,%4,1),%%xmm1,%%xmm1; leaq (%3,%4,2),%3;"\
+  "vaddps %%xmm"#c1",%%xmm1,%%xmm"#c1";"
+
+#define save_c_m8n4(c1,c2,c3,c4)\
+  "vunpcklpd %%ymm"#c2",%%ymm"#c1",%%ymm0; vunpckhpd %%ymm"#c2",%%ymm"#c1",%%ymm1;"\
+  "vunpcklpd %%ymm"#c4",%%ymm"#c3",%%ymm2; vunpckhpd %%ymm"#c4",%%ymm"#c3",%%ymm3;"\
+  "vperm2f128 $2,%%ymm0,%%ymm2,%%ymm"#c1"; vperm2f128 $2,%%ymm1,%%ymm3,%%ymm"#c2";"\
+  "vmovups %%ymm"#c1",(%3); vmovups %%ymm"#c2",(%3,%4,1); leaq (%3,%4,2),%3;"\
+  "vperm2f128 $19,%%ymm0,%%ymm2,%%ymm"#c3"; vperm2f128 $19,%%ymm1,%%ymm3,%%ymm"#c4";"\
+  "vmovups %%ymm"#c3",(%3); vmovups %%ymm"#c4",(%3,%4,1); leaq (%3,%4,2),%3;"
+
+#define save_c_m4n4(c1,c2)\
+  "vunpcklpd %%ymm"#c2",%%ymm"#c1",%%ymm0; vunpckhpd %%ymm"#c2",%%ymm"#c1",%%ymm1;"\
+  "vmovups %%xmm0,(%3); vmovups %%xmm1,(%3,%4,1); leaq (%3,%4,2),%3;"\
+  "vextractf128 $1,%%ymm0,(%3); vextractf128 $1,%%ymm1,(%3,%4,1); leaq (%3,%4,2),%3;"
+
+#define save_c_m2n4(c1)\
+  "vextractf128 $1,%%ymm"#c1",%%xmm1; vmovsd %%xmm"#c1",(%3); vmovhpd %%xmm"#c1",(%3,%4,1); leaq (%3,%4,2),%3;"\
+  "vmovsd %%xmm1,(%3); vmovhpd %%xmm1,(%3,%4,1); leaq (%3,%4,2),%3;"
+
+#define save_c_m1n4(c1)\
+  "vmovss %%xmm"#c1",(%3); vextractps $1,%%xmm"#c1",(%3,%4,1); leaq (%3,%4,2),%3;"\
+  "vextractps $2,%%xmm"#c1",(%3); vextractps $3,%%xmm"#c1",(%3,%4,1); leaq (%3,%4,2),%3;"
+
+#define SOLVE_up_m2n4(a_off,c1)\
+  "vbroadcastsd "#a_off"(%0),%%ymm0; vblendps $170,%8,%%ymm0,%%ymm2;"\
+  "vmulps %%ymm2,%%ymm"#c1",%%ymm"#c1";"\
+  "vmovsldup %%ymm"#c1",%%ymm1;"
+
+#define SOLVE_up_m2n8(a_off,c1,c2)\
+  "vbroadcastsd "#a_off"(%0),%%ymm0; vblendps $170,%8,%%ymm0,%%ymm2;"\
+  "vmulps %%ymm2,%%ymm"#c1",%%ymm"#c1"; vmulps %%ymm2,%%ymm"#c2",%%ymm"#c2";"\
+  "vmovsldup %%ymm"#c1",%%ymm1; vmovsldup %%ymm"#c2",%%ymm2;"
+
+#define SOLVE_up_m2n12(a_off,c1,c2,c3)\
+  "vbroadcastsd "#a_off"(%0),%%ymm0; vblendps $170,%8,%%ymm0,%%ymm2;"\
+  "vmulps %%ymm2,%%ymm"#c1",%%ymm"#c1"; vmulps %%ymm2,%%ymm"#c2",%%ymm"#c2"; vmulps %%ymm2,%%ymm"#c3",%%ymm"#c3";"\
+  "vmovsldup %%ymm"#c1",%%ymm1; vmovsldup %%ymm"#c2",%%ymm2; vmovsldup %%ymm"#c3",%%ymm3;"
+
+#define SOLVE_uplo_m2n4(a_off,c1) SOLVE_up_m2n4(a_off,c1)\
+  "vblendps $85,%9,%%ymm0,%%ymm0; vfnmadd231ps %%ymm0,%%ymm1,%%ymm"#c1";"
+
+#define SOLVE_uplo_m2n8(a_off,c1,c2) SOLVE_up_m2n8(a_off,c1,c2)\
+  "vblendps $85,%9,%%ymm0,%%ymm0; vfnmadd231ps %%ymm0,%%ymm1,%%ymm"#c1"; vfnmadd231ps %%ymm0,%%ymm2,%%ymm"#c2";"
+
+#define SOLVE_uplo_m2n12(a_off,c1,c2,c3) SOLVE_up_m2n12(a_off,c1,c2,c3)\
+  "vblendps $85,%9,%%ymm0,%%ymm0; vfnmadd231ps %%ymm0,%%ymm1,%%ymm"#c1"; vfnmadd231ps %%ymm0,%%ymm2,%%ymm"#c2"; vfnmadd231ps %%ymm0,%%ymm3,%%ymm"#c3";"
+
+#define SOLVE_lo_m2n4(a_off,c1)\
+  "vbroadcastsd "#a_off"(%0),%%ymm0; vblendps $85,%8,%%ymm0,%%ymm2;"\
+  "vmulps %%ymm2,%%ymm"#c1",%%ymm"#c1";"\
+  "vmovshdup %%ymm"#c1",%%ymm1;"
+
+#define SOLVE_lo_m2n8(a_off,c1,c2)\
+  "vbroadcastsd "#a_off"(%0),%%ymm0; vblendps $85,%8,%%ymm0,%%ymm2;"\
+  "vmulps %%ymm2,%%ymm"#c1",%%ymm"#c1"; vmulps %%ymm2,%%ymm"#c2",%%ymm"#c2";"\
+  "vmovshdup %%ymm"#c1",%%ymm1; vmovshdup %%ymm"#c2",%%ymm2;"
+
+#define SOLVE_lo_m2n12(a_off,c1,c2,c3)\
+  "vbroadcastsd "#a_off"(%0),%%ymm0; vblendps $85,%8,%%ymm0,%%ymm2;"\
+  "vmulps %%ymm2,%%ymm"#c1",%%ymm"#c1"; vmulps %%ymm2,%%ymm"#c2",%%ymm"#c2"; vmulps %%ymm2,%%ymm"#c3",%%ymm"#c3";"\
+  "vmovshdup %%ymm"#c1",%%ymm1; vmovshdup %%ymm"#c2",%%ymm2; vmovshdup %%ymm"#c3",%%ymm3;"
+
+#define SOLVE_loup_m2n4(a_off,c1) SOLVE_lo_m2n4(a_off,c1)\
+  "vblendps $170,%9,%%ymm0,%%ymm0; vfnmadd231ps %%ymm0,%%ymm1,%%ymm"#c1";"
+
+#define SOLVE_loup_m2n8(a_off,c1,c2) SOLVE_lo_m2n8(a_off,c1,c2)\
+  "vblendps $170,%9,%%ymm0,%%ymm0; vfnmadd231ps %%ymm0,%%ymm1,%%ymm"#c1"; vfnmadd231ps %%ymm0,%%ymm2,%%ymm"#c2";"
+
+#define SOLVE_loup_m2n12(a_off,c1,c2,c3) SOLVE_lo_m2n12(a_off,c1,c2,c3)\
+  "vblendps $170,%9,%%ymm0,%%ymm0; vfnmadd231ps %%ymm0,%%ymm1,%%ymm"#c1"; vfnmadd231ps %%ymm0,%%ymm2,%%ymm"#c2"; vfnmadd231ps %%ymm0,%%ymm3,%%ymm"#c3";"
+
+#define SOLVE_m1n4(a_off,c1) "vbroadcastss "#a_off"(%0),%%xmm0; vmulps %%xmm0,%%xmm"#c1",%%xmm"#c1";"
+#define SOLVE_m1n8(a_off,c1,c2) SOLVE_m1n4(a_off,c1) "vmulps %%xmm0,%%xmm"#c2",%%xmm"#c2";"
+#define SOLVE_m1n12(a_off,c1,c2,c3) SOLVE_m1n8(a_off,c1,c2) "vmulps %%xmm0,%%xmm"#c3",%%xmm"#c3";"
+
+#define SUBTRACT_m2n4(a_off,c1) "vbroadcastsd "#a_off"(%0),%%ymm0; vfnmadd231ps %%ymm0,%%ymm1,%%ymm"#c1";"
+#define SUBTRACT_m2n8(a_off,c1,c2) SUBTRACT_m2n4(a_off,c1) "vfnmadd231ps %%ymm0,%%ymm2,%%ymm"#c2";"
+#define SUBTRACT_m2n12(a_off,c1,c2,c3) SUBTRACT_m2n8(a_off,c1,c2) "vfnmadd231ps %%ymm0,%%ymm3,%%ymm"#c3";"
+
+#define save_b_m2n4(c1,tmp,b_off,...)\
+  "vpermilps $216,%%ymm"#c1",%%ymm"#tmp"; vpermpd $216,%%ymm"#tmp",%%ymm"#tmp"; vmovups %%ymm"#tmp","#b_off"("#__VA_ARGS__");"
+
+#define SAVE_b_m2n4(b_off,c1) save_b_m2n4(c1,1,b_off,%1)
+#define SAVE_b_m2n8(b_off,c1,c2) SAVE_b_m2n4(b_off,c1) save_b_m2n4(c2,2,b_off,%1,%%r12,4)
+#define SAVE_b_m2n12(b_off,c1,c2,c3) SAVE_b_m2n8(b_off,c1,c2) save_b_m2n4(c3,3,b_off,%1,%%r12,8)
+
+#define SAVE_b_m1n4(b_off,c1) "vmovups %%xmm"#c1","#b_off"(%1);"
+#define SAVE_b_m1n8(b_off,c1,c2) SAVE_b_m1n4(b_off,c1) "vmovups %%xmm"#c2","#b_off"(%1,%%r12,4);"
+#define SAVE_b_m1n12(b_off,c1,c2,c3) SAVE_b_m1n8(b_off,c1,c2) "vmovups %%xmm"#c3","#b_off"(%1,%%r12,8);"
+
diff --git a/kernel/x86_64/strsm_kernel_8x4_haswell_RN.c b/kernel/x86_64/strsm_kernel_8x4_haswell_RN.c
new file mode 100644
index 000000000..4e2cd4fe6
--- /dev/null
+++ b/kernel/x86_64/strsm_kernel_8x4_haswell_RN.c
@@ -0,0 +1,279 @@
+#include "common.h"
+#include <stdint.h>
+#include "strsm_kernel_8x4_haswell_R_common.h"
+
+#define SOLVE_RN_m8n4 \
+  "movq %2,%3;" GEMM_SUM_REORDER_8x4(4,5,6,7,63) "movq %2,%3; addq $32,%2;"\
+  SOLVE_leri_m8n2(0,4,5,%1) SUBTRACT_m8n2(8,6,7,%1)\
+  SOLVE_ri_m8n2(16,4,5,%1) SUBTRACT_m8n2(24,6,7,%1)\
+  SAVE_SOLUTION_m8n2(4,5,0)\
+  SOLVE_leri_m8n2(40,6,7,%1)\
+  SOLVE_ri_m8n2(56,6,7,%1)\
+  SAVE_SOLUTION_m8n2(6,7,64)
+
+#define SOLVE_RN_m8n8 \
+  "movq %2,%3;" GEMM_SUM_REORDER_8x4(4,5,6,7,63) GEMM_SUM_REORDER_8x4(8,9,10,11,63) "movq %2,%3; addq $32,%2;"\
+  SOLVE_leri_m8n2(0,4,5,%1) SUBTRACT_m8n2(8,6,7,%1) SUBTRACT_m8n2(0,8,9,%1,%%r12,4) SUBTRACT_m8n2(8,10,11,%1,%%r12,4)\
+  SOLVE_ri_m8n2(16,4,5,%1) SUBTRACT_m8n2(24,6,7,%1) SUBTRACT_m8n2(16,8,9,%1,%%r12,4) SUBTRACT_m8n2(24,10,11,%1,%%r12,4)\
+  SAVE_SOLUTION_m8n2(4,5,0)\
+  SOLVE_leri_m8n2(40,6,7,%1) SUBTRACT_m8n2(32,8,9,%1,%%r12,4) SUBTRACT_m8n2(40,10,11,%1,%%r12,4)\
+  SOLVE_ri_m8n2(56,6,7,%1) SUBTRACT_m8n2(48,8,9,%1,%%r12,4) SUBTRACT_m8n2(56,10,11,%1,%%r12,4)\
+  SAVE_SOLUTION_m8n2(6,7,64)\
+  SOLVE_leri_m8n2(64,8,9,%1,%%r12,4) SUBTRACT_m8n2(72,10,11,%1,%%r12,4)\
+  SOLVE_ri_m8n2(80,8,9,%1,%%r12,4) SUBTRACT_m8n2(88,10,11,%1,%%r12,4)\
+  SAVE_SOLUTION_m8n2(8,9,128)\
+  SOLVE_leri_m8n2(104,10,11,%1,%%r12,4)\
+  SOLVE_ri_m8n2(120,10,11,%1,%%r12,4)\
+  SAVE_SOLUTION_m8n2(10,11,192)
+
+#define SOLVE_RN_m8n12 \
+  "movq %2,%3;" GEMM_SUM_REORDER_8x4(4,5,6,7,63) GEMM_SUM_REORDER_8x4(8,9,10,11,63) GEMM_SUM_REORDER_8x4(12,13,14,15,63) "movq %2,%3; addq $32,%2;"\
+  SOLVE_leri_m8n2(0,4,5,%1) SUBTRACT_m8n2(8,6,7,%1) SUBTRACT_m8n2(0,8,9,%1,%%r12,4) SUBTRACT_m8n2(8,10,11,%1,%%r12,4) SUBTRACT_m8n2(0,12,13,%1,%%r12,8) SUBTRACT_m8n2(8,14,15,%1,%%r12,8)\
+  SOLVE_ri_m8n2(16,4,5,%1) SUBTRACT_m8n2(24,6,7,%1) SUBTRACT_m8n2(16,8,9,%1,%%r12,4) SUBTRACT_m8n2(24,10,11,%1,%%r12,4) SUBTRACT_m8n2(16,12,13,%1,%%r12,8) SUBTRACT_m8n2(24,14,15,%1,%%r12,8)\
+  SAVE_SOLUTION_m8n2(4,5,0)\
+  SOLVE_leri_m8n2(40,6,7,%1) SUBTRACT_m8n2(32,8,9,%1,%%r12,4) SUBTRACT_m8n2(40,10,11,%1,%%r12,4) SUBTRACT_m8n2(32,12,13,%1,%%r12,8) SUBTRACT_m8n2(40,14,15,%1,%%r12,8)\
+  SOLVE_ri_m8n2(56,6,7,%1) SUBTRACT_m8n2(48,8,9,%1,%%r12,4) SUBTRACT_m8n2(56,10,11,%1,%%r12,4) SUBTRACT_m8n2(48,12,13,%1,%%r12,8) SUBTRACT_m8n2(56,14,15,%1,%%r12,8)\
+  SAVE_SOLUTION_m8n2(6,7,64)\
+  SOLVE_leri_m8n2(64,8,9,%1,%%r12,4) SUBTRACT_m8n2(72,10,11,%1,%%r12,4) SUBTRACT_m8n2(64,12,13,%1,%%r12,8) SUBTRACT_m8n2(72,14,15,%1,%%r12,8)\
+  SOLVE_ri_m8n2(80,8,9,%1,%%r12,4) SUBTRACT_m8n2(88,10,11,%1,%%r12,4) SUBTRACT_m8n2(80,12,13,%1,%%r12,8) SUBTRACT_m8n2(88,14,15,%1,%%r12,8)\
+  SAVE_SOLUTION_m8n2(8,9,128)\
+  SOLVE_leri_m8n2(104,10,11,%1,%%r12,4) SUBTRACT_m8n2(96,12,13,%1,%%r12,8) SUBTRACT_m8n2(104,14,15,%1,%%r12,8)\
+  SOLVE_ri_m8n2(120,10,11,%1,%%r12,4) SUBTRACT_m8n2(112,12,13,%1,%%r12,8) SUBTRACT_m8n2(120,14,15,%1,%%r12,8)\
+  SAVE_SOLUTION_m8n2(10,11,192)\
+  SOLVE_leri_m8n2(128,12,13,%1,%%r12,8) SUBTRACT_m8n2(136,14,15,%1,%%r12,8)\
+  SOLVE_ri_m8n2(144,12,13,%1,%%r12,8) SUBTRACT_m8n2(152,14,15,%1,%%r12,8)\
+  SAVE_SOLUTION_m8n2(12,13,256)\
+  SOLVE_leri_m8n2(168,14,15,%1,%%r12,8)\
+  SOLVE_ri_m8n2(184,14,15,%1,%%r12,8)\
+  SAVE_SOLUTION_m8n2(14,15,320)
+
+#define SOLVE_RN_m4n4 \
+  "movq %2,%3;" GEMM_SUM_REORDER_4x4(4,5,6,7,4,5) "movq %2,%3; addq $16,%2;"\
+  SOLVE_leri_m4n2(0,4,%1) SUBTRACT_m4n2(8,5,%1)\
+  SOLVE_ri_m4n2(16,4,%1) SUBTRACT_m4n2(24,5,%1)\
+  SAVE_SOLUTION_m4n2(4,0)\
+  SOLVE_leri_m4n2(40,5,%1)\
+  SOLVE_ri_m4n2(56,5,%1)\
+  SAVE_SOLUTION_m4n2(5,32)
+
+#define SOLVE_RN_m4n8 \
+  "movq %2,%3;" GEMM_SUM_REORDER_4x4(4,5,6,7,4,5) GEMM_SUM_REORDER_4x4(8,9,10,11,6,7) "movq %2,%3; addq $16,%2;"\
+  SOLVE_leri_m4n2(0,4,%1) SUBTRACT_m4n2(8,5,%1) SUBTRACT_m4n2(0,6,%1,%%r12,4) SUBTRACT_m4n2(8,7,%1,%%r12,4)\
+  SOLVE_ri_m4n2(16,4,%1) SUBTRACT_m4n2(24,5,%1) SUBTRACT_m4n2(16,6,%1,%%r12,4) SUBTRACT_m4n2(24,7,%1,%%r12,4)\
+  SAVE_SOLUTION_m4n2(4,0)\
+  SOLVE_leri_m4n2(40,5,%1) SUBTRACT_m4n2(32,6,%1,%%r12,4) SUBTRACT_m4n2(40,7,%1,%%r12,4)\
+  SOLVE_ri_m4n2(56,5,%1) SUBTRACT_m4n2(48,6,%1,%%r12,4) SUBTRACT_m4n2(56,7,%1,%%r12,4)\
+  SAVE_SOLUTION_m4n2(5,32)\
+  SOLVE_leri_m4n2(64,6,%1,%%r12,4) SUBTRACT_m4n2(72,7,%1,%%r12,4)\
+  SOLVE_ri_m4n2(80,6,%1,%%r12,4) SUBTRACT_m4n2(88,7,%1,%%r12,4)\
+  SAVE_SOLUTION_m4n2(6,64)\
+  SOLVE_leri_m4n2(104,7,%1,%%r12,4)\
+  SOLVE_ri_m4n2(120,7,%1,%%r12,4)\
+  SAVE_SOLUTION_m4n2(7,96)
+
+#define SOLVE_RN_m4n12 \
+  "movq %2,%3;" GEMM_SUM_REORDER_4x4(4,5,6,7,4,5) GEMM_SUM_REORDER_4x4(8,9,10,11,6,7) GEMM_SUM_REORDER_4x4(12,13,14,15,8,9) "movq %2,%3; addq $16,%2;"\
+  SOLVE_leri_m4n2(0,4,%1) SUBTRACT_m4n2(8,5,%1) SUBTRACT_m4n2(0,6,%1,%%r12,4) SUBTRACT_m4n2(8,7,%1,%%r12,4) SUBTRACT_m4n2(0,8,%1,%%r12,8) SUBTRACT_m4n2(8,9,%1,%%r12,8)\
+  SOLVE_ri_m4n2(16,4,%1) SUBTRACT_m4n2(24,5,%1) SUBTRACT_m4n2(16,6,%1,%%r12,4) SUBTRACT_m4n2(24,7,%1,%%r12,4) SUBTRACT_m4n2(16,8,%1,%%r12,8) SUBTRACT_m4n2(24,9,%1,%%r12,8)\
+  SAVE_SOLUTION_m4n2(4,0)\
+  SOLVE_leri_m4n2(40,5,%1) SUBTRACT_m4n2(32,6,%1,%%r12,4) SUBTRACT_m4n2(40,7,%1,%%r12,4) SUBTRACT_m4n2(32,8,%1,%%r12,8) SUBTRACT_m4n2(40,9,%1,%%r12,8)\
+  SOLVE_ri_m4n2(56,5,%1) SUBTRACT_m4n2(48,6,%1,%%r12,4) SUBTRACT_m4n2(56,7,%1,%%r12,4) SUBTRACT_m4n2(48,8,%1,%%r12,8) SUBTRACT_m4n2(56,9,%1,%%r12,8)\
+  SAVE_SOLUTION_m4n2(5,32)\
+  SOLVE_leri_m4n2(64,6,%1,%%r12,4) SUBTRACT_m4n2(72,7,%1,%%r12,4) SUBTRACT_m4n2(64,8,%1,%%r12,8) SUBTRACT_m4n2(72,9,%1,%%r12,8)\
+  SOLVE_ri_m4n2(80,6,%1,%%r12,4) SUBTRACT_m4n2(88,7,%1,%%r12,4) SUBTRACT_m4n2(80,8,%1,%%r12,8) SUBTRACT_m4n2(88,9,%1,%%r12,8)\
+  SAVE_SOLUTION_m4n2(6,64)\
+  SOLVE_leri_m4n2(104,7,%1,%%r12,4) SUBTRACT_m4n2(96,8,%1,%%r12,8) SUBTRACT_m4n2(104,9,%1,%%r12,8)\
+  SOLVE_ri_m4n2(120,7,%1,%%r12,4) SUBTRACT_m4n2(112,8,%1,%%r12,8) SUBTRACT_m4n2(120,9,%1,%%r12,8)\
+  SAVE_SOLUTION_m4n2(7,96)\
+  SOLVE_leri_m4n2(128,8,%1,%%r12,8) SUBTRACT_m4n2(136,9,%1,%%r12,8)\
+  SOLVE_ri_m4n2(144,8,%1,%%r12,8) SUBTRACT_m4n2(152,9,%1,%%r12,8)\
+  SAVE_SOLUTION_m4n2(8,128)\
+  SOLVE_leri_m4n2(168,9,%1,%%r12,8)\
+  SOLVE_ri_m4n2(184,9,%1,%%r12,8)\
+  SAVE_SOLUTION_m4n2(9,160)
+
+#define SOLVE_RN_m2n4 \
+  "movq %2,%3;" GEMM_SUM_REORDER_2x4(4,5) "movq %2,%3; addq $8,%2;"\
+  SOLVE_col1_ltor_m2n4(0,4,5,%1)\
+  SOLVE_col2_ltor_m2n4(16,4,5,%1)\
+  SOLVE_col3_ltor_m2n4(32,4,5,%1)\
+  SOLVE_col4_ltor_m2n4(48,4,5,%1)\
+  SAVE_SOLUTION_m2n4(4,5,0)
+
+#define SOLVE_RN_m2n8 \
+  "movq %2,%3;" GEMM_SUM_REORDER_2x4(4,5) GEMM_SUM_REORDER_2x4(6,7) "movq %2,%3; addq $8,%2;"\
+  SOLVE_col1_ltor_m2n4(0,4,5,%1) SUBTRACT_m2n4(0,6,7,%1,%%r12,4)\
+  SOLVE_col2_ltor_m2n4(16,4,5,%1) SUBTRACT_m2n4(16,6,7,%1,%%r12,4)\
+  SOLVE_col3_ltor_m2n4(32,4,5,%1) SUBTRACT_m2n4(32,6,7,%1,%%r12,4)\
+  SOLVE_col4_ltor_m2n4(48,4,5,%1) SUBTRACT_m2n4(48,6,7,%1,%%r12,4)\
+  SAVE_SOLUTION_m2n4(4,5,0)\
+  SOLVE_col1_ltor_m2n4(64,6,7,%1,%%r12,4)\
+  SOLVE_col2_ltor_m2n4(80,6,7,%1,%%r12,4)\
+  SOLVE_col3_ltor_m2n4(96,6,7,%1,%%r12,4)\
+  SOLVE_col4_ltor_m2n4(112,6,7,%1,%%r12,4)\
+  SAVE_SOLUTION_m2n4(6,7,32)
+
+#define SOLVE_RN_m2n12 \
+  "movq %2,%3;" GEMM_SUM_REORDER_2x4(4,5) GEMM_SUM_REORDER_2x4(6,7) GEMM_SUM_REORDER_2x4(8,9) "movq %2,%3; addq $8,%2;"\
+  SOLVE_col1_ltor_m2n4(0,4,5,%1) SUBTRACT_m2n4(0,6,7,%1,%%r12,4) SUBTRACT_m2n4(0,8,9,%1,%%r12,8)\
+  SOLVE_col2_ltor_m2n4(16,4,5,%1) SUBTRACT_m2n4(16,6,7,%1,%%r12,4) SUBTRACT_m2n4(16,8,9,%1,%%r12,8)\
+  SOLVE_col3_ltor_m2n4(32,4,5,%1) SUBTRACT_m2n4(32,6,7,%1,%%r12,4) SUBTRACT_m2n4(32,8,9,%1,%%r12,8)\
+  SOLVE_col4_ltor_m2n4(48,4,5,%1) SUBTRACT_m2n4(48,6,7,%1,%%r12,4) SUBTRACT_m2n4(48,8,9,%1,%%r12,8)\
+  SAVE_SOLUTION_m2n4(4,5,0)\
+  SOLVE_col1_ltor_m2n4(64,6,7,%1,%%r12,4) SUBTRACT_m2n4(64,8,9,%1,%%r12,8)\
+  SOLVE_col2_ltor_m2n4(80,6,7,%1,%%r12,4) SUBTRACT_m2n4(80,8,9,%1,%%r12,8)\
+  SOLVE_col3_ltor_m2n4(96,6,7,%1,%%r12,4) SUBTRACT_m2n4(96,8,9,%1,%%r12,8)\
+  SOLVE_col4_ltor_m2n4(112,6,7,%1,%%r12,4) SUBTRACT_m2n4(112,8,9,%1,%%r12,8)\
+  SAVE_SOLUTION_m2n4(6,7,32)\
+  SOLVE_col1_ltor_m2n4(128,8,9,%1,%%r12,8)\
+  SOLVE_col2_ltor_m2n4(144,8,9,%1,%%r12,8)\
+  SOLVE_col3_ltor_m2n4(160,8,9,%1,%%r12,8)\
+  SOLVE_col4_ltor_m2n4(176,8,9,%1,%%r12,8)\
+  SAVE_SOLUTION_m2n4(8,9,64)
+
+#define SOLVE_RN_m1n4 \
+  "movq %2,%3;" GEMM_SUM_REORDER_1x4(4) "movq %2,%3; addq $4,%2;"\
+  SOLVE_col1_ltor_m1n4(0,4,%1)\
+  SOLVE_col2_ltor_m1n4(16,4,%1)\
+  SOLVE_col3_ltor_m1n4(32,4,%1)\
+  SOLVE_col4_ltor_m1n4(48,4,%1)\
+  SAVE_SOLUTION_m1n4(4,0)
+
+#define SOLVE_RN_m1n8 \
+  "movq %2,%3;" GEMM_SUM_REORDER_1x4(4) GEMM_SUM_REORDER_1x4(5) "movq %2,%3; addq $4,%2;"\
+  SOLVE_col1_ltor_m1n4(0,4,%1) SUBTRACT_m1n4(0,5,%1,%%r12,4)\
+  SOLVE_col2_ltor_m1n4(16,4,%1) SUBTRACT_m1n4(16,5,%1,%%r12,4)\
+  SOLVE_col3_ltor_m1n4(32,4,%1) SUBTRACT_m1n4(32,5,%1,%%r12,4)\
+  SOLVE_col4_ltor_m1n4(48,4,%1) SUBTRACT_m1n4(48,5,%1,%%r12,4)\
+  SAVE_SOLUTION_m1n4(4,0)\
+  SOLVE_col1_ltor_m1n4(64,5,%1,%%r12,4)\
+  SOLVE_col2_ltor_m1n4(80,5,%1,%%r12,4)\
+  SOLVE_col3_ltor_m1n4(96,5,%1,%%r12,4)\
+  SOLVE_col4_ltor_m1n4(112,5,%1,%%r12,4)\
+  SAVE_SOLUTION_m1n4(5,16)
+
+#define SOLVE_RN_m1n12 \
+  "movq %2,%3;" GEMM_SUM_REORDER_1x4(4) GEMM_SUM_REORDER_1x4(5) GEMM_SUM_REORDER_1x4(6) "movq %2,%3; addq $4,%2;"\
+  SOLVE_col1_ltor_m1n4(0,4,%1) SUBTRACT_m1n4(0,5,%1,%%r12,4) SUBTRACT_m1n4(0,6,%1,%%r12,8)\
+  SOLVE_col2_ltor_m1n4(16,4,%1) SUBTRACT_m1n4(16,5,%1,%%r12,4) SUBTRACT_m1n4(16,6,%1,%%r12,8)\
+  SOLVE_col3_ltor_m1n4(32,4,%1) SUBTRACT_m1n4(32,5,%1,%%r12,4) SUBTRACT_m1n4(32,6,%1,%%r12,8)\
+  SOLVE_col4_ltor_m1n4(48,4,%1) SUBTRACT_m1n4(48,5,%1,%%r12,4) SUBTRACT_m1n4(48,6,%1,%%r12,8)\
+  SAVE_SOLUTION_m1n4(4,0)\
+  SOLVE_col1_ltor_m1n4(64,5,%1,%%r12,4) SUBTRACT_m1n4(64,6,%1,%%r12,8)\
+  SOLVE_col2_ltor_m1n4(80,5,%1,%%r12,4) SUBTRACT_m1n4(80,6,%1,%%r12,8)\
+  SOLVE_col3_ltor_m1n4(96,5,%1,%%r12,4) SUBTRACT_m1n4(96,6,%1,%%r12,8)\
+  SOLVE_col4_ltor_m1n4(112,5,%1,%%r12,4) SUBTRACT_m1n4(112,6,%1,%%r12,8)\
+  SAVE_SOLUTION_m1n4(5,16)\
+  SOLVE_col1_ltor_m1n4(128,6,%1,%%r12,8)\
+  SOLVE_col2_ltor_m1n4(144,6,%1,%%r12,8)\
+  SOLVE_col3_ltor_m1n4(160,6,%1,%%r12,8)\
+  SOLVE_col4_ltor_m1n4(176,6,%1,%%r12,8)\
+  SAVE_SOLUTION_m1n4(6,32)
+
+#define GEMM_RN_SIMPLE(mdim,ndim) \
+  "movq %%r15,%0; leaq (%%r15,%%r12,"#mdim"),%%r15; movq %%r13,%5; movq %%r14,%1;" INIT_m##mdim##n##ndim\
+  "testq %5,%5; jz 1"#mdim""#ndim"2f;"\
+  "1"#mdim""#ndim"1:\n\t"\
+  GEMM_KERNEL_k1m##mdim##n##ndim "addq $16,%1; addq $"#mdim"*4,%0; decq %5; jnz 1"#mdim""#ndim"1b;"\
+  "1"#mdim""#ndim"2:\n\t"
+#define GEMM_RN_m8n4 GEMM_RN_SIMPLE(8,4)
+#define GEMM_RN_m8n8 GEMM_RN_SIMPLE(8,8)
+#define GEMM_RN_m8n12 \
+  "movq %%r15,%0; leaq (%%r15,%%r12,8),%%r15; movq %%r13,%5; movq %%r14,%1;" INIT_m8n12\
+  "cmpq $8,%5; jb 18122f;"\
+  "18121:\n\t"\
+  GEMM_KERNEL_k1m8n12 "prefetcht0 384(%0); addq $32,%0; addq $16,%1;"\
+  GEMM_KERNEL_k1m8n12 "addq $32,%0; addq $16,%1;"\
+  GEMM_KERNEL_k1m8n12 "prefetcht0 384(%0); addq $32,%0; addq $16,%1;"\
+  GEMM_KERNEL_k1m8n12 "addq $32,%0; addq $16,%1;"\
+  GEMM_KERNEL_k1m8n12 "prefetcht0 384(%0); addq $32,%0; addq $16,%1;"\
+  GEMM_KERNEL_k1m8n12 "addq $32,%0; addq $16,%1;"\
+  GEMM_KERNEL_k1m8n12 "prefetcht0 384(%0); addq $32,%0; addq $16,%1;"\
+  GEMM_KERNEL_k1m8n12 "addq $32,%0; addq $16,%1;"\
+  "subq $8,%5; cmpq $8,%5; jnb 18121b;"\
+  "18122:\n\t"\
+  "testq %5,%5; jz 18124f;"\
+  "18123:\n\t"\
+  GEMM_KERNEL_k1m8n12 "addq $32,%0; addq $16,%1; decq %5; jnz 18123b;"\
+  "18124:\n\t"
+#define GEMM_RN_m4n4 GEMM_RN_SIMPLE(4,4)
+#define GEMM_RN_m4n8 GEMM_RN_SIMPLE(4,8)
+#define GEMM_RN_m4n12 GEMM_RN_SIMPLE(4,12)
+#define GEMM_RN_m2n4 GEMM_RN_SIMPLE(2,4)
+#define GEMM_RN_m2n8 GEMM_RN_SIMPLE(2,8)
+#define GEMM_RN_m2n12 GEMM_RN_SIMPLE(2,12)
+#define GEMM_RN_m1n4 GEMM_RN_SIMPLE(1,4)
+#define GEMM_RN_m1n8 GEMM_RN_SIMPLE(1,8)
+#define GEMM_RN_m1n12 GEMM_RN_SIMPLE(1,12)
+
+#define COMPUTE(ndim) {\
+  __asm__ __volatile__(\
+    "movq %0,%%r15; movq %1,%%r14; movq %7,%%r13; movq %6,%%r12; salq $2,%%r12; movq %10,%%r11;"\
+    "cmpq $8,%%r11; jb "#ndim"772f;"\
+    #ndim"771:\n\t"\
+    GEMM_RN_m8n##ndim SOLVE_RN_m8n##ndim "subq $8,%%r11; cmpq $8,%%r11; jnb "#ndim"771b;"\
+    #ndim"772:\n\t"\
+    "testq $4,%%r11; jz "#ndim"773f;"\
+    GEMM_RN_m4n##ndim SOLVE_RN_m4n##ndim "subq $4,%%r11;"\
+    #ndim"773:\n\t"\
+    "testq $2,%%r11; jz "#ndim"774f;"\
+    GEMM_RN_m2n##ndim SOLVE_RN_m2n##ndim "subq $2,%%r11;"\
+    #ndim"774:\n\t"\
+    "testq $1,%%r11; jz "#ndim"775f;"\
+    GEMM_RN_m1n##ndim SOLVE_RN_m1n##ndim "subq $1,%%r11;"\
+    #ndim"775:\n\t"\
+    "movq %%r15,%0; movq %%r14,%1; vzeroupper;"\
+  :"+r"(a_ptr),"+r"(b_ptr),"+r"(c_ptr),"+r"(c_tmp),"+r"(ldc_bytes),"+r"(k_cnt):"m"(K),"m"(OFF),"m"(one[0]),"m"(zero[0]),"m"(M)\
+  :"r11","r12","r13","r14","r15","cc","memory",\
+  "xmm0","xmm1","xmm2","xmm3","xmm4","xmm5","xmm6","xmm7","xmm8","xmm9","xmm10","xmm11","xmm12","xmm13","xmm14","xmm15");\
+  a_ptr -= M * K; b_ptr += ndim * K; c_ptr += ldc * ndim - M; OFF += ndim;\
+}
+
+static void solve_RN(BLASLONG m, BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc) {
+  FLOAT a0, b0;
+  int i, j, k;
+  for (i=0; i<n; i++) {
+    b0 = b[i*n+i];
+    for (j=0; j<m; j++) {
+      a0 = c[i*ldc+j] * b0;
+      a[i*m+j] = c[i*ldc+j] = a0;
+      for (k=i+1; k<n; k++) c[k*ldc+j] -= a0 * b[i*n+k];
+    }
+  }
+}
+static void COMPUTE_EDGE_1_nchunk(BLASLONG m, BLASLONG n, FLOAT *sa, FLOAT *sb, FLOAT *C, BLASLONG ldc, BLASLONG k, BLASLONG offset) {
+  BLASLONG m_count = m, kk = offset; FLOAT *a_ptr = sa, *c_ptr = C;
+  for(;m_count>7;m_count-=8){
+    if(kk>0) GEMM_KERNEL_N(8,n,kk,-1.0,a_ptr,sb,c_ptr,ldc);
+    solve_RN(8,n,a_ptr+kk*8,sb+kk*n,c_ptr,ldc);
+    a_ptr += k * 8; c_ptr += 8;
+  }
+  for(;m_count>3;m_count-=4){
+    if(kk>0) GEMM_KERNEL_N(4,n,kk,-1.0,a_ptr,sb,c_ptr,ldc);
+    solve_RN(4,n,a_ptr+kk*4,sb+kk*n,c_ptr,ldc);
+    a_ptr += k * 4; c_ptr += 4;
+  }
+  for(;m_count>1;m_count-=2){
+    if(kk>0) GEMM_KERNEL_N(2,n,kk,-1.0,a_ptr,sb,c_ptr,ldc);
+    solve_RN(2,n,a_ptr+kk*2,sb+kk*n,c_ptr,ldc);
+    a_ptr += k * 2; c_ptr += 2;
+  }
+  if(m_count>0){
+    if(kk>0) GEMM_KERNEL_N(1,n,kk,-1.0,a_ptr,sb,c_ptr,ldc);
+    solve_RN(1,n,a_ptr+kk*1,sb+kk*n,c_ptr,ldc);
+    a_ptr += k * 1; c_ptr += 1;
+  }
+}
+int CNAME(BLASLONG m, BLASLONG n, BLASLONG k, FLOAT dummy1, FLOAT *sa, FLOAT *sb, FLOAT *C, BLASLONG ldc, BLASLONG offset){
+  float *a_ptr = sa, *b_ptr = sb, *c_ptr = C, *c_tmp = C;
+  float one[8] = {1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0};
+  float zero[8] = {0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0};
+  uint64_t ldc_bytes = (uint64_t)ldc * sizeof(float), K = (uint64_t)k, M = (uint64_t)m, OFF = (uint64_t)-offset, k_cnt = 0;
+  BLASLONG n_count = n;
+  for(;n_count>11;n_count-=12) COMPUTE(12)
+  for(;n_count>7;n_count-=8) COMPUTE(8)
+  for(;n_count>3;n_count-=4) COMPUTE(4)
+  for(;n_count>1;n_count-=2) { COMPUTE_EDGE_1_nchunk(m,2,a_ptr,b_ptr,c_ptr,ldc,k,OFF); b_ptr += 2*k; c_ptr += ldc*2; OFF+=2;}
+  if(n_count>0) COMPUTE_EDGE_1_nchunk(m,1,a_ptr,b_ptr,c_ptr,ldc,k,OFF);
+  return 0;
+}
diff --git a/kernel/x86_64/strsm_kernel_8x4_haswell_RT.c b/kernel/x86_64/strsm_kernel_8x4_haswell_RT.c
new file mode 100644
index 000000000..ffcbfbbf0
--- /dev/null
+++ b/kernel/x86_64/strsm_kernel_8x4_haswell_RT.c
@@ -0,0 +1,281 @@
+#include "common.h"
+#include <stdint.h>
+#include "strsm_kernel_8x4_haswell_R_common.h"
+
+#define SOLVE_RT_m8n4 \
+  "movq %2,%3;" GEMM_SUM_REORDER_8x4(4,5,6,7,63) "negq %4; leaq (%3,%4,2),%3; negq %4; addq $32,%2;"\
+  SOLVE_rile_m8n2(-8,6,7,%1) SUBTRACT_m8n2(-16,4,5,%1)\
+  SOLVE_le_m8n2(-24,6,7,%1) SUBTRACT_m8n2(-32,4,5,%1)\
+  SAVE_SOLUTION_m8n2(6,7,-64) "negq %4; leaq (%3,%4,4),%3; negq %4;"\
+  SOLVE_rile_m8n2(-48,4,5,%1)\
+  SOLVE_le_m8n2(-64,4,5,%1)\
+  SAVE_SOLUTION_m8n2(4,5,-128)
+
+#define SOLVE_RT_m8n8 \
+  "movq %2,%3;" GEMM_SUM_REORDER_8x4(4,5,6,7,63) GEMM_SUM_REORDER_8x4(8,9,10,11,63) "negq %4; leaq (%3,%4,2),%3; negq %4; addq $32,%2;"\
+  SOLVE_rile_m8n2(-8,10,11,%1,%%r12,4) SUBTRACT_m8n2(-16,8,9,%1,%%r12,4) SUBTRACT_m8n2(-8,6,7,%1) SUBTRACT_m8n2(-16,4,5,%1)\
+  SOLVE_le_m8n2(-24,10,11,%1,%%r12,4) SUBTRACT_m8n2(-32,8,9,%1,%%r12,4) SUBTRACT_m8n2(-24,6,7,%1) SUBTRACT_m8n2(-32,4,5,%1)\
+  SAVE_SOLUTION_m8n2(10,11,-64) "negq %4; leaq (%3,%4,4),%3; negq %4;"\
+  SOLVE_rile_m8n2(-48,8,9,%1,%%r12,4) SUBTRACT_m8n2(-40,6,7,%1) SUBTRACT_m8n2(-48,4,5,%1)\
+  SOLVE_le_m8n2(-64,8,9,%1,%%r12,4) SUBTRACT_m8n2(-56,6,7,%1) SUBTRACT_m8n2(-64,4,5,%1)\
+  SAVE_SOLUTION_m8n2(8,9,-128) "negq %4; leaq (%3,%4,4),%3; negq %4;"\
+  SOLVE_rile_m8n2(-72,6,7,%1) SUBTRACT_m8n2(-80,4,5,%1)\
+  SOLVE_le_m8n2(-88,6,7,%1) SUBTRACT_m8n2(-96,4,5,%1)\
+  SAVE_SOLUTION_m8n2(6,7,-192) "negq %4; leaq (%3,%4,4),%3; negq %4;"\
+  SOLVE_rile_m8n2(-112,4,5,%1)\
+  SOLVE_le_m8n2(-128,4,5,%1)\
+  SAVE_SOLUTION_m8n2(4,5,-256)
+
+#define SOLVE_RT_m8n12 \
+  "movq %2,%3;" GEMM_SUM_REORDER_8x4(4,5,6,7,63) GEMM_SUM_REORDER_8x4(8,9,10,11,63) GEMM_SUM_REORDER_8x4(12,13,14,15,63) "negq %4; leaq (%3,%4,2),%3; negq %4; addq $32,%2;"\
+  SOLVE_rile_m8n2(-8,14,15,%1,%%r12,8) SUBTRACT_m8n2(-16,12,13,%1,%%r12,8) SUBTRACT_m8n2(-8,10,11,%1,%%r12,4) SUBTRACT_m8n2(-16,8,9,%1,%%r12,4) SUBTRACT_m8n2(-8,6,7,%1) SUBTRACT_m8n2(-16,4,5,%1)\
+  SOLVE_le_m8n2(-24,14,15,%1,%%r12,8) SUBTRACT_m8n2(-32,12,13,%1,%%r12,8) SUBTRACT_m8n2(-24,10,11,%1,%%r12,4) SUBTRACT_m8n2(-32,8,9,%1,%%r12,4) SUBTRACT_m8n2(-24,6,7,%1) SUBTRACT_m8n2(-32,4,5,%1)\
+  SAVE_SOLUTION_m8n2(14,15,-64) "negq %4; leaq (%3,%4,4),%3; negq %4;"\
+  SOLVE_rile_m8n2(-48,12,13,%1,%%r12,8) SUBTRACT_m8n2(-40,10,11,%1,%%r12,4) SUBTRACT_m8n2(-48,8,9,%1,%%r12,4) SUBTRACT_m8n2(-40,6,7,%1) SUBTRACT_m8n2(-48,4,5,%1)\
+  SOLVE_le_m8n2(-64,12,13,%1,%%r12,8) SUBTRACT_m8n2(-56,10,11,%1,%%r12,4) SUBTRACT_m8n2(-64,8,9,%1,%%r12,4) SUBTRACT_m8n2(-56,6,7,%1) SUBTRACT_m8n2(-64,4,5,%1)\
+  SAVE_SOLUTION_m8n2(12,13,-128) "negq %4; leaq (%3,%4,4),%3; negq %4;"\
+  SOLVE_rile_m8n2(-72,10,11,%1,%%r12,4) SUBTRACT_m8n2(-80,8,9,%1,%%r12,4) SUBTRACT_m8n2(-72,6,7,%1) SUBTRACT_m8n2(-80,4,5,%1)\
+  SOLVE_le_m8n2(-88,10,11,%1,%%r12,4) SUBTRACT_m8n2(-96,8,9,%1,%%r12,4) SUBTRACT_m8n2(-88,6,7,%1) SUBTRACT_m8n2(-96,4,5,%1)\
+  SAVE_SOLUTION_m8n2(10,11,-192) "negq %4; leaq (%3,%4,4),%3; negq %4;"\
+  SOLVE_rile_m8n2(-112,8,9,%1,%%r12,4) SUBTRACT_m8n2(-104,6,7,%1) SUBTRACT_m8n2(-112,4,5,%1)\
+  SOLVE_le_m8n2(-128,8,9,%1,%%r12,4) SUBTRACT_m8n2(-120,6,7,%1) SUBTRACT_m8n2(-128,4,5,%1)\
+  SAVE_SOLUTION_m8n2(8,9,-256) "negq %4; leaq (%3,%4,4),%3; negq %4;"\
+  SOLVE_rile_m8n2(-136,6,7,%1) SUBTRACT_m8n2(-144,4,5,%1)\
+  SOLVE_le_m8n2(-152,6,7,%1) SUBTRACT_m8n2(-160,4,5,%1)\
+  SAVE_SOLUTION_m8n2(6,7,-320) "negq %4; leaq (%3,%4,4),%3; negq %4;"\
+  SOLVE_rile_m8n2(-176,4,5,%1)\
+  SOLVE_le_m8n2(-192,4,5,%1)\
+  SAVE_SOLUTION_m8n2(4,5,-384)
+
+#define SOLVE_RT_m4n4 \
+  "movq %2,%3;" GEMM_SUM_REORDER_4x4(4,5,6,7,4,5) "negq %4; leaq (%3,%4,2),%3; negq %4; addq $16,%2;"\
+  SOLVE_rile_m4n2(-8,5,%1) SUBTRACT_m4n2(-16,4,%1)\
+  SOLVE_le_m4n2(-24,5,%1) SUBTRACT_m4n2(-32,4,%1)\
+  SAVE_SOLUTION_m4n2(5,-32) "negq %4; leaq (%3,%4,4),%3; negq %4;"\
+  SOLVE_rile_m4n2(-48,4,%1)\
+  SOLVE_le_m4n2(-64,4,%1)\
+  SAVE_SOLUTION_m4n2(4,-64)
+
+#define SOLVE_RT_m4n8 \
+  "movq %2,%3;" GEMM_SUM_REORDER_4x4(4,5,6,7,4,5) GEMM_SUM_REORDER_4x4(8,9,10,11,6,7) "negq %4; leaq (%3,%4,2),%3; negq %4; addq $16,%2;"\
+  SOLVE_rile_m4n2(-8,7,%1,%%r12,4) SUBTRACT_m4n2(-16,6,%1,%%r12,4) SUBTRACT_m4n2(-8,5,%1) SUBTRACT_m4n2(-16,4,%1)\
+  SOLVE_le_m4n2(-24,7,%1,%%r12,4) SUBTRACT_m4n2(-32,6,%1,%%r12,4) SUBTRACT_m4n2(-24,5,%1) SUBTRACT_m4n2(-32,4,%1)\
+  SAVE_SOLUTION_m4n2(7,-32) "negq %4; leaq (%3,%4,4),%3; negq %4;"\
+  SOLVE_rile_m4n2(-48,6,%1,%%r12,4) SUBTRACT_m4n2(-40,5,%1) SUBTRACT_m4n2(-48,4,%1)\
+  SOLVE_le_m4n2(-64,6,%1,%%r12,4) SUBTRACT_m4n2(-56,5,%1) SUBTRACT_m4n2(-64,4,%1)\
+  SAVE_SOLUTION_m4n2(6,-64) "negq %4; leaq (%3,%4,4),%3; negq %4;"\
+  SOLVE_rile_m4n2(-72,5,%1) SUBTRACT_m4n2(-80,4,%1)\
+  SOLVE_le_m4n2(-88,5,%1) SUBTRACT_m4n2(-96,4,%1)\
+  SAVE_SOLUTION_m4n2(5,-96) "negq %4; leaq (%3,%4,4),%3; negq %4;"\
+  SOLVE_rile_m4n2(-112,4,%1)\
+  SOLVE_le_m4n2(-128,4,%1)\
+  SAVE_SOLUTION_m4n2(4,-128)
+
+#define SOLVE_RT_m4n12 \
+  "movq %2,%3;" GEMM_SUM_REORDER_4x4(4,5,6,7,4,5) GEMM_SUM_REORDER_4x4(8,9,10,11,6,7) GEMM_SUM_REORDER_4x4(12,13,14,15,8,9) "negq %4; leaq (%3,%4,2),%3; negq %4; addq $16,%2;"\
+  SOLVE_rile_m4n2(-8,9,%1,%%r12,8) SUBTRACT_m4n2(-16,8,%1,%%r12,8) SUBTRACT_m4n2(-8,7,%1,%%r12,4) SUBTRACT_m4n2(-16,6,%1,%%r12,4) SUBTRACT_m4n2(-8,5,%1) SUBTRACT_m4n2(-16,4,%1)\
+  SOLVE_le_m4n2(-24,9,%1,%%r12,8) SUBTRACT_m4n2(-32,8,%1,%%r12,8) SUBTRACT_m4n2(-24,7,%1,%%r12,4) SUBTRACT_m4n2(-32,6,%1,%%r12,4) SUBTRACT_m4n2(-24,5,%1) SUBTRACT_m4n2(-32,4,%1)\
+  SAVE_SOLUTION_m4n2(9,-32) "negq %4; leaq (%3,%4,4),%3; negq %4;"\
+  SOLVE_rile_m4n2(-48,8,%1,%%r12,8) SUBTRACT_m4n2(-40,7,%1,%%r12,4) SUBTRACT_m4n2(-48,6,%1,%%r12,4) SUBTRACT_m4n2(-40,5,%1) SUBTRACT_m4n2(-48,4,%1)\
+  SOLVE_le_m4n2(-64,8,%1,%%r12,8) SUBTRACT_m4n2(-56,7,%1,%%r12,4) SUBTRACT_m4n2(-64,6,%1,%%r12,4) SUBTRACT_m4n2(-56,5,%1) SUBTRACT_m4n2(-64,4,%1)\
+  SAVE_SOLUTION_m4n2(8,-64) "negq %4; leaq (%3,%4,4),%3; negq %4;"\
+  SOLVE_rile_m4n2(-72,7,%1,%%r12,4) SUBTRACT_m4n2(-80,6,%1,%%r12,4) SUBTRACT_m4n2(-72,5,%1) SUBTRACT_m4n2(-80,4,%1)\
+  SOLVE_le_m4n2(-88,7,%1,%%r12,4) SUBTRACT_m4n2(-96,6,%1,%%r12,4) SUBTRACT_m4n2(-88,5,%1) SUBTRACT_m4n2(-96,4,%1)\
+  SAVE_SOLUTION_m4n2(7,-96) "negq %4; leaq (%3,%4,4),%3; negq %4;"\
+  SOLVE_rile_m4n2(-112,6,%1,%%r12,4) SUBTRACT_m4n2(-104,5,%1) SUBTRACT_m4n2(-112,4,%1)\
+  SOLVE_le_m4n2(-128,6,%1,%%r12,4) SUBTRACT_m4n2(-120,5,%1) SUBTRACT_m4n2(-128,4,%1)\
+  SAVE_SOLUTION_m4n2(6,-128) "negq %4; leaq (%3,%4,4),%3; negq %4;"\
+  SOLVE_rile_m4n2(-136,5,%1) SUBTRACT_m4n2(-144,4,%1)\
+  SOLVE_le_m4n2(-152,5,%1) SUBTRACT_m4n2(-160,4,%1)\
+  SAVE_SOLUTION_m4n2(5,-160) "negq %4; leaq (%3,%4,4),%3; negq %4;"\
+  SOLVE_rile_m4n2(-176,4,%1)\
+  SOLVE_le_m4n2(-192,4,%1)\
+  SAVE_SOLUTION_m4n2(4,-192)
+
+#define SOLVE_RT_m2n4 \
+  "movq %2,%3;" GEMM_SUM_REORDER_2x4(4,5) "negq %4; leaq (%3,%4,4),%3; negq %4; addq $8,%2;"\
+  SOLVE_col4_rtol_m2n4(-16,4,5,%1)\
+  SOLVE_col3_rtol_m2n4(-32,4,5,%1)\
+  SOLVE_col2_rtol_m2n4(-48,4,5,%1)\
+  SOLVE_col1_rtol_m2n4(-64,4,5,%1)\
+  SAVE_SOLUTION_m2n4(4,5,-32)
+
+#define SOLVE_RT_m2n8 \
+  "movq %2,%3;" GEMM_SUM_REORDER_2x4(4,5) GEMM_SUM_REORDER_2x4(6,7) "negq %4; leaq (%3,%4,4),%3; negq %4; addq $8,%2;"\
+  SOLVE_col4_rtol_m2n4(-16,6,7,%1,%%r12,4) SUBTRACT_m2n4(-16,4,5,%1)\
+  SOLVE_col3_rtol_m2n4(-32,6,7,%1,%%r12,4) SUBTRACT_m2n4(-32,4,5,%1)\
+  SOLVE_col2_rtol_m2n4(-48,6,7,%1,%%r12,4) SUBTRACT_m2n4(-48,4,5,%1)\
+  SOLVE_col1_rtol_m2n4(-64,6,7,%1,%%r12,4) SUBTRACT_m2n4(-64,4,5,%1)\
+  SAVE_SOLUTION_m2n4(6,7,-32) "negq %4; leaq (%3,%4,8),%3; negq %4;"\
+  SOLVE_col4_rtol_m2n4(-80,4,5,%1)\
+  SOLVE_col3_rtol_m2n4(-96,4,5,%1)\
+  SOLVE_col2_rtol_m2n4(-112,4,5,%1)\
+  SOLVE_col1_rtol_m2n4(-128,4,5,%1)\
+  SAVE_SOLUTION_m2n4(4,5,-64)
+
+#define SOLVE_RT_m2n12 \
+  "movq %2,%3;" GEMM_SUM_REORDER_2x4(4,5) GEMM_SUM_REORDER_2x4(6,7) GEMM_SUM_REORDER_2x4(8,9) "negq %4; leaq (%3,%4,4),%3; negq %4; addq $8,%2;"\
+  SOLVE_col4_rtol_m2n4(-16,8,9,%1,%%r12,8) SUBTRACT_m2n4(-16,6,7,%1,%%r12,4) SUBTRACT_m2n4(-16,4,5,%1)\
+  SOLVE_col3_rtol_m2n4(-32,8,9,%1,%%r12,8) SUBTRACT_m2n4(-32,6,7,%1,%%r12,4) SUBTRACT_m2n4(-32,4,5,%1)\
+  SOLVE_col2_rtol_m2n4(-48,8,9,%1,%%r12,8) SUBTRACT_m2n4(-48,6,7,%1,%%r12,4) SUBTRACT_m2n4(-48,4,5,%1)\
+  SOLVE_col1_rtol_m2n4(-64,8,9,%1,%%r12,8) SUBTRACT_m2n4(-64,6,7,%1,%%r12,4) SUBTRACT_m2n4(-64,4,5,%1)\
+  SAVE_SOLUTION_m2n4(8,9,-32) "negq %4; leaq (%3,%4,8),%3; negq %4;"\
+  SOLVE_col4_rtol_m2n4(-80,6,7,%1,%%r12,4) SUBTRACT_m2n4(-80,4,5,%1)\
+  SOLVE_col3_rtol_m2n4(-96,6,7,%1,%%r12,4) SUBTRACT_m2n4(-96,4,5,%1)\
+  SOLVE_col2_rtol_m2n4(-112,6,7,%1,%%r12,4) SUBTRACT_m2n4(-112,4,5,%1)\
+  SOLVE_col1_rtol_m2n4(-128,6,7,%1,%%r12,4) SUBTRACT_m2n4(-128,4,5,%1)\
+  SAVE_SOLUTION_m2n4(6,7,-64) "negq %4; leaq (%3,%4,8),%3; negq %4;"\
+  SOLVE_col4_rtol_m2n4(-144,4,5,%1)\
+  SOLVE_col3_rtol_m2n4(-160,4,5,%1)\
+  SOLVE_col2_rtol_m2n4(-176,4,5,%1)\
+  SOLVE_col1_rtol_m2n4(-192,4,5,%1)\
+  SAVE_SOLUTION_m2n4(4,5,-96)
+
+#define SOLVE_RT_m1n4 \
+  "movq %2,%3;" GEMM_SUM_REORDER_1x4(4) "negq %4; leaq (%3,%4,4),%3; negq %4; addq $4,%2;"\
+  SOLVE_col4_rtol_m1n4(-16,4,%1)\
+  SOLVE_col3_rtol_m1n4(-32,4,%1)\
+  SOLVE_col2_rtol_m1n4(-48,4,%1)\
+  SOLVE_col1_rtol_m1n4(-64,4,%1)\
+  SAVE_SOLUTION_m1n4(4,-16)
+
+#define SOLVE_RT_m1n8 \
+  "movq %2,%3;" GEMM_SUM_REORDER_1x4(4) GEMM_SUM_REORDER_1x4(5) "negq %4; leaq (%3,%4,4),%3; negq %4; addq $4,%2;"\
+  SOLVE_col4_rtol_m1n4(-16,5,%1,%%r12,4) SUBTRACT_m1n4(-16,4,%1)\
+  SOLVE_col3_rtol_m1n4(-32,5,%1,%%r12,4) SUBTRACT_m1n4(-32,4,%1)\
+  SOLVE_col2_rtol_m1n4(-48,5,%1,%%r12,4) SUBTRACT_m1n4(-48,4,%1)\
+  SOLVE_col1_rtol_m1n4(-64,5,%1,%%r12,4) SUBTRACT_m1n4(-64,4,%1)\
+  SAVE_SOLUTION_m1n4(5,-16) "negq %4; leaq (%3,%4,8),%3; negq %4;"\
+  SOLVE_col4_rtol_m1n4(-80,4,%1)\
+  SOLVE_col3_rtol_m1n4(-96,4,%1)\
+  SOLVE_col2_rtol_m1n4(-112,4,%1)\
+  SOLVE_col1_rtol_m1n4(-128,4,%1)\
+  SAVE_SOLUTION_m1n4(4,-32)
+
+#define SOLVE_RT_m1n12 \
+  "movq %2,%3;" GEMM_SUM_REORDER_1x4(4) GEMM_SUM_REORDER_1x4(5) GEMM_SUM_REORDER_1x4(6) "negq %4; leaq (%3,%4,4),%3; negq %4; addq $4,%2;"\
+  SOLVE_col4_rtol_m1n4(-16,6,%1,%%r12,8) SUBTRACT_m1n4(-16,5,%1,%%r12,4) SUBTRACT_m1n4(-16,4,%1)\
+  SOLVE_col3_rtol_m1n4(-32,6,%1,%%r12,8) SUBTRACT_m1n4(-32,5,%1,%%r12,4) SUBTRACT_m1n4(-32,4,%1)\
+  SOLVE_col2_rtol_m1n4(-48,6,%1,%%r12,8) SUBTRACT_m1n4(-48,5,%1,%%r12,4) SUBTRACT_m1n4(-48,4,%1)\
+  SOLVE_col1_rtol_m1n4(-64,6,%1,%%r12,8) SUBTRACT_m1n4(-64,5,%1,%%r12,4) SUBTRACT_m1n4(-64,4,%1)\
+  SAVE_SOLUTION_m1n4(6,-16) "negq %4; leaq (%3,%4,8),%3; negq %4;"\
+  SOLVE_col4_rtol_m1n4(-80,5,%1,%%r12,4) SUBTRACT_m1n4(-80,4,%1)\
+  SOLVE_col3_rtol_m1n4(-96,5,%1,%%r12,4) SUBTRACT_m1n4(-96,4,%1)\
+  SOLVE_col2_rtol_m1n4(-112,5,%1,%%r12,4) SUBTRACT_m1n4(-112,4,%1)\
+  SOLVE_col1_rtol_m1n4(-128,5,%1,%%r12,4) SUBTRACT_m1n4(-128,4,%1)\
+  SAVE_SOLUTION_m1n4(5,-32) "negq %4; leaq (%3,%4,8),%3; negq %4;"\
+  SOLVE_col4_rtol_m1n4(-144,4,%1)\
+  SOLVE_col3_rtol_m1n4(-160,4,%1)\
+  SOLVE_col2_rtol_m1n4(-176,4,%1)\
+  SOLVE_col1_rtol_m1n4(-192,4,%1)\
+  SAVE_SOLUTION_m1n4(4,-48)
+
+/* r14 = b_tail, r15 = a_tail, r13 = k-kk */
+#define GEMM_RT_SIMPLE(mdim,ndim) \
+  "leaq (%%r15,%%r12,"#mdim"),%%r15; movq %%r15,%0; movq %%r13,%5; movq %%r14,%1;" INIT_m##mdim##n##ndim\
+  "testq %5,%5; jz 1"#mdim""#ndim"2f;"\
+  "1"#mdim""#ndim"1:\n\t"\
+  "subq $16,%1; subq $"#mdim"*4,%0;" GEMM_KERNEL_k1m##mdim##n##ndim "decq %5; jnz 1"#mdim""#ndim"1b;"\
+  "1"#mdim""#ndim"2:\n\t"
+#define GEMM_RT_m8n4 GEMM_RT_SIMPLE(8,4)
+#define GEMM_RT_m8n8 GEMM_RT_SIMPLE(8,8)
+#define GEMM_RT_m8n12 \
+  "leaq (%%r15,%%r12,8),%%r15; movq %%r15,%0; movq %%r13,%5; movq %%r14,%1;" INIT_m8n12\
+  "cmpq $8,%5; jb 18122f;"\
+  "18121:\n\t"\
+  "prefetcht0 -384(%0); subq $32,%0; subq $16,%1;" GEMM_KERNEL_k1m8n12\
+                       "subq $32,%0; subq $16,%1;" GEMM_KERNEL_k1m8n12\
+  "prefetcht0 -384(%0); subq $32,%0; subq $16,%1;" GEMM_KERNEL_k1m8n12\
+                       "subq $32,%0; subq $16,%1;" GEMM_KERNEL_k1m8n12\
+  "prefetcht0 -384(%0); subq $32,%0; subq $16,%1;" GEMM_KERNEL_k1m8n12\
+                       "subq $32,%0; subq $16,%1;" GEMM_KERNEL_k1m8n12\
+  "prefetcht0 -384(%0); subq $32,%0; subq $16,%1;" GEMM_KERNEL_k1m8n12\
+                       "subq $32,%0; subq $16,%1;" GEMM_KERNEL_k1m8n12\
+  "subq $8,%5; cmpq $8,%5; jnb 18121b;"\
+  "18122:\n\t"\
+  "testq %5,%5; jz 18124f;"\
+  "18123:\n\t"\
+  "subq $32,%0; subq $16,%1;" GEMM_KERNEL_k1m8n12 "decq %5; jnz 18123b;"\
+  "18124:\n\t"
+#define GEMM_RT_m4n4 GEMM_RT_SIMPLE(4,4)
+#define GEMM_RT_m4n8 GEMM_RT_SIMPLE(4,8)
+#define GEMM_RT_m4n12 GEMM_RT_SIMPLE(4,12)
+#define GEMM_RT_m2n4 GEMM_RT_SIMPLE(2,4)
+#define GEMM_RT_m2n8 GEMM_RT_SIMPLE(2,8)
+#define GEMM_RT_m2n12 GEMM_RT_SIMPLE(2,12)
+#define GEMM_RT_m1n4 GEMM_RT_SIMPLE(1,4)
+#define GEMM_RT_m1n8 GEMM_RT_SIMPLE(1,8)
+#define GEMM_RT_m1n12 GEMM_RT_SIMPLE(1,12)
+
+#define COMPUTE(ndim) {\
+  b_ptr -= (ndim-4)*K; c_ptr -= ndim * ldc;\
+  __asm__ __volatile__(\
+    "movq %0,%%r15; movq %6,%%r13; subq %7,%%r13; movq %6,%%r12; salq $2,%%r12; movq %1,%%r14; movq %10,%%r11;"\
+    "cmpq $8,%%r11; jb "#ndim"772f;"\
+    #ndim"771:\n\t"\
+    GEMM_RT_m8n##ndim SOLVE_RT_m8n##ndim "subq $8,%%r11; cmpq $8,%%r11; jnb "#ndim"771b;"\
+    #ndim"772:\n\t"\
+    "testq $4,%%r11; jz "#ndim"773f;"\
+    GEMM_RT_m4n##ndim SOLVE_RT_m4n##ndim "subq $4,%%r11;"\
+    #ndim"773:\n\t"\
+    "testq $2,%%r11; jz "#ndim"774f;"\
+    GEMM_RT_m2n##ndim SOLVE_RT_m2n##ndim "subq $2,%%r11;"\
+    #ndim"774:\n\t"\
+    "testq $1,%%r11; jz "#ndim"775f;"\
+    GEMM_RT_m1n##ndim SOLVE_RT_m1n##ndim "subq $1,%%r11;"\
+    #ndim"775:\n\t"\
+    "movq %%r15,%0; movq %%r14,%1; vzeroupper;"\
+  :"+r"(a_ptr),"+r"(b_ptr),"+r"(c_ptr),"+r"(c_tmp),"+r"(ldc_bytes),"+r"(k_cnt):"m"(K),"m"(OFF),"m"(one[0]),"m"(zero[0]),"m"(M)\
+  :"r11","r12","r13","r14","r15","cc","memory",\
+  "xmm0","xmm1","xmm2","xmm3","xmm4","xmm5","xmm6","xmm7","xmm8","xmm9","xmm10","xmm11","xmm12","xmm13","xmm14","xmm15");\
+  a_ptr -= M * K; b_ptr -= 4 * K; c_ptr -= M; OFF -= ndim;\
+}
+
+static void solve_RT(BLASLONG m, BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc){
+  FLOAT a0, b0;
+  int i, j, k;
+  for (i=n-1;i>=0;i--) {
+    b0 = b[i*n+i];
+    for (j=0;j<m;j++) {
+      a0 = c[i*ldc+j] * b0;
+      a[i*m+j] = c[i*ldc+j] = a0;
+      for (k=0;k<i;k++) c[k*ldc+j] -= a0 * b[i*n+k];
+    }
+  }
+}
+static void COMPUTE_EDGE_1_nchunk(BLASLONG m, BLASLONG n, FLOAT *sa, FLOAT *sb, FLOAT *C, BLASLONG ldc, BLASLONG k, BLASLONG offset) {
+  BLASLONG m_count = m, kk = offset; FLOAT *a_ptr = sa, *c_ptr = C;
+  for(;m_count>7;m_count-=8){
+    if(k-kk>0) GEMM_KERNEL_N(8,n,k-kk,-1.0,a_ptr+kk*8,sb+kk*n,c_ptr,ldc);
+    solve_RT(8,n,a_ptr+(kk-n)*8,sb+(kk-n)*n,c_ptr,ldc);
+    a_ptr += k * 8; c_ptr += 8;
+  }
+  for(;m_count>3;m_count-=4){
+    if(k-kk>0) GEMM_KERNEL_N(4,n,k-kk,-1.0,a_ptr+kk*4,sb+kk*n,c_ptr,ldc);
+    solve_RT(4,n,a_ptr+(kk-n)*4,sb+(kk-n)*n,c_ptr,ldc);
+    a_ptr += k * 4; c_ptr += 4;
+  }
+  for(;m_count>1;m_count-=2){
+    if(k-kk>0) GEMM_KERNEL_N(2,n,k-kk,-1.0,a_ptr+kk*2,sb+kk*n,c_ptr,ldc);
+    solve_RT(2,n,a_ptr+(kk-n)*2,sb+(kk-n)*n,c_ptr,ldc);
+    a_ptr += k * 2; c_ptr += 2;
+  }
+  if(m_count>0){
+    if(k-kk>0) GEMM_KERNEL_N(1,n,k-kk,-1.0,a_ptr+kk*1,sb+kk*n,c_ptr,ldc);
+    solve_RT(1,n,a_ptr+(kk-n)*1,sb+(kk-n)*n,c_ptr,ldc);
+    a_ptr += k * 1; c_ptr += 1;
+  }
+}
+int CNAME(BLASLONG m, BLASLONG n, BLASLONG k, FLOAT dummy1, FLOAT *sa, FLOAT *sb, FLOAT *C, BLASLONG ldc, BLASLONG offset){
+  float *a_ptr = sa, *b_ptr = sb+n*k, *c_ptr = C+n*ldc, *c_tmp = C;
+  float one[8] = {1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0};
+  float zero[8] = {0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0};
+  uint64_t ldc_bytes = (uint64_t)ldc * sizeof(float), K = (uint64_t)k, M = (uint64_t)m, OFF = (uint64_t)(n-offset), k_cnt = 0;
+  BLASLONG n_count = n;
+  if(n&1){b_ptr-=k; c_ptr-=ldc; COMPUTE_EDGE_1_nchunk(m,1,a_ptr,b_ptr,c_ptr,ldc,k,OFF); OFF--; n_count--;}
+  if(n&2){b_ptr-=k*2; c_ptr-=ldc*2; COMPUTE_EDGE_1_nchunk(m,2,a_ptr,b_ptr,c_ptr,ldc,k,OFF); OFF-=2; n_count-=2;}
+  for(;n_count>11;n_count-=12) COMPUTE(12)
+  for(;n_count>7;n_count-=8) COMPUTE(8)
+  for(;n_count>3;n_count-=4) COMPUTE(4)
+  return 0;
+}
diff --git a/kernel/x86_64/strsm_kernel_8x4_haswell_R_common.h b/kernel/x86_64/strsm_kernel_8x4_haswell_R_common.h
new file mode 100644
index 000000000..36b7aa1a3
--- /dev/null
+++ b/kernel/x86_64/strsm_kernel_8x4_haswell_R_common.h
@@ -0,0 +1,226 @@
+/* r11 = m_counter, r12 = size_of_k_elements, r13 = kk, r14 = b_head, r15 = a_head */
+/* register i/o: %0 = a_ptr, %1 = b_ptr, %2 = c_ptr, %3 = c_tmp, %4 = ldc, %5 = k_counter */
+/* memory input: %6 = K, %7 = offset, %8 = {1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0}, %9 = {0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0}, %10 = M */
+
+#define init_m8n4(c1,c2,c3,c4)\
+  "vpxor %%ymm"#c1",%%ymm"#c1",%%ymm"#c1"; vpxor %%ymm"#c2",%%ymm"#c2",%%ymm"#c2";"\
+  "vpxor %%ymm"#c3",%%ymm"#c3",%%ymm"#c3"; vpxor %%ymm"#c4",%%ymm"#c4",%%ymm"#c4";"
+#define INIT_m8n4 init_m8n4(4,5,6,7)
+#define INIT_m8n8 INIT_m8n4 init_m8n4(8,9,10,11)
+#define INIT_m8n12 INIT_m8n8 init_m8n4(12,13,14,15)
+
+#define init_m4n4(c1,c2,c3,c4)\
+  "vpxor %%xmm"#c1",%%xmm"#c1",%%xmm"#c1"; vpxor %%xmm"#c2",%%xmm"#c2",%%xmm"#c2";"\
+  "vpxor %%xmm"#c3",%%xmm"#c3",%%xmm"#c3"; vpxor %%xmm"#c4",%%xmm"#c4",%%xmm"#c4";"
+#define INIT_m4n4 init_m4n4(4,5,6,7)
+#define INIT_m4n8 INIT_m4n4 init_m4n4(8,9,10,11)
+#define INIT_m4n12 INIT_m4n8 init_m4n4(12,13,14,15)
+
+#define init_m2n4(c1,c2)\
+  "vpxor %%xmm"#c1",%%xmm"#c1",%%xmm"#c1"; vpxor %%xmm"#c2",%%xmm"#c2",%%xmm"#c2";"
+#define INIT_m2n4 init_m2n4(4,5)
+#define INIT_m2n8 INIT_m2n4 init_m2n4(6,7)
+#define INIT_m2n12 INIT_m2n8 init_m2n4(8,9)
+
+#define init_m1n4(c1) "vpxor %%xmm"#c1",%%xmm"#c1",%%xmm"#c1";"
+#define INIT_m1n4 init_m1n4(4)
+#define INIT_m1n8 INIT_m1n4 init_m1n4(5)
+#define INIT_m1n12 INIT_m1n8 init_m1n4(6)
+
+#define GEMM_KERNEL_k1m8n4 \
+  "vmovsldup (%0),%%ymm1; vmovshdup (%0),%%ymm2;"\
+  "vbroadcastsd (%1),%%ymm3; vfnmadd231ps %%ymm3,%%ymm1,%%ymm4; vfnmadd231ps %%ymm3,%%ymm2,%%ymm5;"\
+  "vbroadcastsd 8(%1),%%ymm3; vfnmadd231ps %%ymm3,%%ymm1,%%ymm6; vfnmadd231ps %%ymm3,%%ymm2,%%ymm7;"
+#define GEMM_KERNEL_k1m8n8 GEMM_KERNEL_k1m8n4\
+  "vbroadcastsd (%1,%%r12,4),%%ymm3; vfnmadd231ps %%ymm3,%%ymm1,%%ymm8; vfnmadd231ps %%ymm3,%%ymm2,%%ymm9;"\
+  "vbroadcastsd 8(%1,%%r12,4),%%ymm3; vfnmadd231ps %%ymm3,%%ymm1,%%ymm10; vfnmadd231ps %%ymm3,%%ymm2,%%ymm11;"
+#define GEMM_KERNEL_k1m8n12 GEMM_KERNEL_k1m8n8\
+  "vbroadcastsd (%1,%%r12,8),%%ymm3; vfnmadd231ps %%ymm3,%%ymm1,%%ymm12; vfnmadd231ps %%ymm3,%%ymm2,%%ymm13;"\
+  "vbroadcastsd 8(%1,%%r12,8),%%ymm3; vfnmadd231ps %%ymm3,%%ymm1,%%ymm14; vfnmadd231ps %%ymm3,%%ymm2,%%ymm15;"
+
+#define GEMM_KERNEL_k1m4n4 \
+  "vmovsldup (%0),%%xmm1; vmovshdup (%0),%%xmm2;"\
+  "vmovddup (%1),%%xmm3; vfnmadd231ps %%xmm3,%%xmm1,%%xmm4; vfnmadd231ps %%xmm3,%%xmm2,%%xmm5;"\
+  "vmovddup 8(%1),%%xmm3; vfnmadd231ps %%xmm3,%%xmm1,%%xmm6; vfnmadd231ps %%xmm3,%%xmm2,%%xmm7;"
+#define GEMM_KERNEL_k1m4n8 GEMM_KERNEL_k1m4n4\
+  "vmovddup (%1,%%r12,4),%%xmm3; vfnmadd231ps %%xmm3,%%xmm1,%%xmm8; vfnmadd231ps %%xmm3,%%xmm2,%%xmm9;"\
+  "vmovddup 8(%1,%%r12,4),%%xmm3; vfnmadd231ps %%xmm3,%%xmm1,%%xmm10; vfnmadd231ps %%xmm3,%%xmm2,%%xmm11;"
+#define GEMM_KERNEL_k1m4n12 GEMM_KERNEL_k1m4n8\
+  "vmovddup (%1,%%r12,8),%%xmm3; vfnmadd231ps %%xmm3,%%xmm1,%%xmm12; vfnmadd231ps %%xmm3,%%xmm2,%%xmm13;"\
+  "vmovddup 8(%1,%%r12,8),%%xmm3; vfnmadd231ps %%xmm3,%%xmm1,%%xmm14; vfnmadd231ps %%xmm3,%%xmm2,%%xmm15;"
+
+#define GEMM_KERNEL_k1m2n4 \
+  "vbroadcastss (%0),%%xmm1; vbroadcastss 4(%0),%%xmm2;"\
+  "vmovups (%1),%%xmm3; vfnmadd231ps %%xmm3,%%xmm1,%%xmm4; vfnmadd231ps %%xmm3,%%xmm2,%%xmm5;"
+#define GEMM_KERNEL_k1m2n8 GEMM_KERNEL_k1m2n4\
+  "vmovups (%1,%%r12,4),%%xmm3; vfnmadd231ps %%xmm3,%%xmm1,%%xmm6; vfnmadd231ps %%xmm3,%%xmm2,%%xmm7;"
+#define GEMM_KERNEL_k1m2n12 GEMM_KERNEL_k1m2n8\
+  "vmovups (%1,%%r12,8),%%xmm3; vfnmadd231ps %%xmm3,%%xmm1,%%xmm8; vfnmadd231ps %%xmm3,%%xmm2,%%xmm9;"
+
+#define GEMM_KERNEL_k1m1n4 "vbroadcastss (%0),%%xmm1; vfnmadd231ps (%1),%%xmm1,%%xmm4;"
+#define GEMM_KERNEL_k1m1n8 GEMM_KERNEL_k1m1n4 "vfnmadd231ps (%1,%%r12,4),%%xmm1,%%xmm5;"
+#define GEMM_KERNEL_k1m1n12 GEMM_KERNEL_k1m1n8 "vfnmadd231ps (%1,%%r12,8),%%xmm1,%%xmm6;"
+
+#define GEMM_SUM_REORDER_8x4(c1,c2,c3,c4,prefpos)\
+  "vmovups (%3),%%ymm0; vmovups (%3,%4,1),%%ymm1; prefetcht1 "#prefpos"(%3); prefetcht1 "#prefpos"(%3,%4,1); leaq (%3,%4,2),%3;"\
+  "vunpcklps %%ymm1,%%ymm0,%%ymm2; vunpckhps %%ymm1,%%ymm0,%%ymm3; vunpcklpd %%ymm3,%%ymm2,%%ymm0; vunpckhpd %%ymm3,%%ymm2,%%ymm1;"\
+  "vaddps %%ymm0,%%ymm"#c1",%%ymm"#c1"; vaddps %%ymm1,%%ymm"#c2",%%ymm"#c2";"\
+  "vmovups (%3),%%ymm0; vmovups (%3,%4,1),%%ymm1; prefetcht1 "#prefpos"(%3); prefetcht1 "#prefpos"(%3,%4,1); leaq (%3,%4,2),%3;"\
+  "vunpcklps %%ymm1,%%ymm0,%%ymm2; vunpckhps %%ymm1,%%ymm0,%%ymm3; vunpcklpd %%ymm3,%%ymm2,%%ymm0; vunpckhpd %%ymm3,%%ymm2,%%ymm1;"\
+  "vaddps %%ymm0,%%ymm"#c3",%%ymm"#c3"; vaddps %%ymm1,%%ymm"#c4",%%ymm"#c4";"
+
+#define GEMM_SUM_REORDER_4x4(c1,c2,c3,c4,co1,co2)\
+  "vmovups (%3),%%xmm0; vmovups (%3,%4,1),%%xmm1; leaq (%3,%4,2),%3;"\
+  "vunpcklps %%xmm1,%%xmm0,%%xmm2; vunpckhps %%xmm1,%%xmm0,%%xmm3;"\
+  "vunpcklpd %%xmm"#c2",%%xmm"#c1",%%xmm0; vunpckhpd %%xmm"#c2",%%xmm"#c1",%%xmm1;"\
+  "vaddps %%xmm0,%%xmm2,%%xmm"#c1"; vaddps %%xmm1,%%xmm3,%%xmm"#c2";"\
+  "vmovups (%3),%%xmm0; vmovups (%3,%4,1),%%xmm1; leaq (%3,%4,2),%3;"\
+  "vunpcklps %%xmm1,%%xmm0,%%xmm2; vunpckhps %%xmm1,%%xmm0,%%xmm3;"\
+  "vunpcklpd %%xmm"#c4",%%xmm"#c3",%%xmm0; vunpckhpd %%xmm"#c4",%%xmm"#c3",%%xmm1;"\
+  "vaddps %%xmm0,%%xmm2,%%xmm"#c3"; vaddps %%xmm1,%%xmm3,%%xmm"#c4";"\
+  "vperm2f128 $2,%%ymm"#c1",%%ymm"#c2",%%ymm"#co1"; vperm2f128 $2,%%ymm"#c3",%%ymm"#c4",%%ymm"#co2";"
+
+#define GEMM_SUM_REORDER_2x4(c1,c2)\
+  "vmovsd (%3),%%xmm0; vmovhpd (%3,%4,1),%%xmm0,%%xmm0; leaq (%3,%4,2),%3; vpermilps $216,%%xmm0,%%xmm0;"\
+  "vmovsd (%3),%%xmm1; vmovhpd (%3,%4,1),%%xmm1,%%xmm1; leaq (%3,%4,2),%3; vpermilps $216,%%xmm1,%%xmm1;"\
+  "vunpcklpd %%xmm1,%%xmm0,%%xmm2; vaddps %%xmm2,%%xmm"#c1",%%xmm"#c1";"\
+  "vunpckhpd %%xmm1,%%xmm0,%%xmm3; vaddps %%xmm3,%%xmm"#c2",%%xmm"#c2";"\
+
+#define GEMM_SUM_REORDER_1x4(c1)\
+  "vmovss (%3),%%xmm1; vinsertps $16,(%3,%4,1),%%xmm1,%%xmm1; leaq (%3,%4,2),%3;"\
+  "vinsertps $32,(%3),%%xmm1,%%xmm1; vinsertps $48,(%3,%4,1),%%xmm1,%%xmm1; leaq (%3,%4,2),%3;"\
+  "vaddps %%xmm"#c1",%%xmm1,%%xmm"#c1";"
+
+#define SOLVE_le_m4n2(b_off,c1,...)\
+  "vbroadcastsd "#b_off"("#__VA_ARGS__"),%%ymm0; vblendps $170,%8,%%ymm0,%%ymm2;"\
+  "vmulps %%ymm2,%%ymm"#c1",%%ymm"#c1";"\
+  "vmovsldup %%ymm"#c1",%%ymm1;"
+
+#define SOLVE_le_m8n2(b_off,c1,c2,...)\
+  "vbroadcastsd "#b_off"("#__VA_ARGS__"),%%ymm0; vblendps $170,%8,%%ymm0,%%ymm2;"\
+  "vmulps %%ymm2,%%ymm"#c1",%%ymm"#c1"; vmulps %%ymm2,%%ymm"#c2",%%ymm"#c2";"\
+  "vmovsldup %%ymm"#c1",%%ymm1; vmovsldup %%ymm"#c2",%%ymm2;"
+
+#define SOLVE_leri_m4n2(b_off,c1,...) SOLVE_le_m4n2(b_off,c1,__VA_ARGS__)\
+  "vblendps $85,%9,%%ymm0,%%ymm0; vfnmadd231ps %%ymm0,%%ymm1,%%ymm"#c1";"
+
+#define SOLVE_leri_m8n2(b_off,c1,c2,...) SOLVE_le_m8n2(b_off,c1,c2,__VA_ARGS__)\
+  "vblendps $85,%9,%%ymm0,%%ymm0; vfnmadd231ps %%ymm0,%%ymm1,%%ymm"#c1"; vfnmadd231ps %%ymm0,%%ymm2,%%ymm"#c2";"
+
+#define SOLVE_ri_m4n2(b_off,c1,...)\
+  "vbroadcastsd "#b_off"("#__VA_ARGS__"),%%ymm0; vblendps $85,%8,%%ymm0,%%ymm2;"\
+  "vmulps %%ymm2,%%ymm"#c1",%%ymm"#c1";"\
+  "vmovshdup %%ymm"#c1",%%ymm1;"
+
+#define SOLVE_ri_m8n2(b_off,c1,c2,...)\
+  "vbroadcastsd "#b_off"("#__VA_ARGS__"),%%ymm0; vblendps $85,%8,%%ymm0,%%ymm2;"\
+  "vmulps %%ymm2,%%ymm"#c1",%%ymm"#c1"; vmulps %%ymm2,%%ymm"#c2",%%ymm"#c2";"\
+  "vmovshdup %%ymm"#c1",%%ymm1; vmovshdup %%ymm"#c2",%%ymm2;"
+
+#define SOLVE_rile_m4n2(b_off,c1,...) SOLVE_ri_m4n2(b_off,c1,__VA_ARGS__)\
+  "vblendps $170,%9,%%ymm0,%%ymm0; vfnmadd231ps %%ymm0,%%ymm1,%%ymm"#c1";"
+
+#define SOLVE_rile_m8n2(b_off,c1,c2,...) SOLVE_ri_m8n2(b_off,c1,c2,__VA_ARGS__)\
+  "vblendps $170,%9,%%ymm0,%%ymm0; vfnmadd231ps %%ymm0,%%ymm1,%%ymm"#c1"; vfnmadd231ps %%ymm0,%%ymm2,%%ymm"#c2";"
+
+#define SOLVE_col1_rtol_m1n4(b_off,c1,...)\
+  "vmovups "#b_off"("#__VA_ARGS__"),%%xmm0; vblendps $14,%8,%%xmm0,%%xmm2;"\
+  "vmulps %%xmm2,%%xmm"#c1",%%xmm"#c1";"\
+  "vpermilps $0,%%xmm"#c1",%%xmm1;"
+
+#define SOLVE_col1_rtol_m2n4(b_off,c1,c2,...)\
+  "vmovups "#b_off"("#__VA_ARGS__"),%%xmm0; vblendps $14,%8,%%xmm0,%%xmm2;"\
+  "vmulps %%xmm2,%%xmm"#c1",%%xmm"#c1"; vmulps %%xmm2,%%xmm"#c2",%%xmm"#c2";"\
+  "vpermilps $0,%%xmm"#c1",%%xmm1; vpermilps $0,%%xmm"#c2",%%xmm2;"
+
+#define SOLVE_col1_ltor_m1n4(b_off,c1,...) SOLVE_col1_rtol_m1n4(b_off,c1,__VA_ARGS__)\
+  "vblendps $1,%9,%%xmm0,%%xmm0; vfnmadd231ps %%xmm0,%%xmm1,%%xmm"#c1";"
+
+#define SOLVE_col1_ltor_m2n4(b_off,c1,c2,...) SOLVE_col1_rtol_m2n4(b_off,c1,c2,__VA_ARGS__)\
+  "vblendps $1,%9,%%xmm0,%%xmm0; vfnmadd231ps %%xmm0,%%xmm1,%%xmm"#c1"; vfnmadd231ps %%xmm0,%%xmm2,%%xmm"#c2";"
+
+#define SOLVE_col2_mul_m1n4(b_off,c1,...)\
+  "vmovups "#b_off"("#__VA_ARGS__"),%%xmm0; vblendps $13,%8,%%xmm0,%%xmm2;"\
+  "vmulps %%xmm2,%%xmm"#c1",%%xmm"#c1";"\
+  "vpermilps $85,%%xmm"#c1",%%xmm1;"
+
+#define SOLVE_col2_mul_m2n4(b_off,c1,c2,...)\
+  "vmovups "#b_off"("#__VA_ARGS__"),%%xmm0; vblendps $13,%8,%%xmm0,%%xmm2;"\
+  "vmulps %%xmm2,%%xmm"#c1",%%xmm"#c1"; vmulps %%xmm2,%%xmm"#c2",%%xmm"#c2";"\
+  "vpermilps $85,%%xmm"#c1",%%xmm1; vpermilps $85,%%xmm"#c2",%%xmm2;"
+
+#define SOLVE_col2_rtol_m1n4(b_off,c1,...) SOLVE_col2_mul_m1n4(b_off,c1,__VA_ARGS__)\
+  "vblendps $14,%9,%%xmm0,%%xmm0; vfnmadd231ps %%xmm0,%%xmm1,%%xmm"#c1";"
+
+#define SOLVE_col2_rtol_m2n4(b_off,c1,c2,...) SOLVE_col2_mul_m2n4(b_off,c1,c2,__VA_ARGS__)\
+  "vblendps $14,%9,%%xmm0,%%xmm0; vfnmadd231ps %%xmm0,%%xmm1,%%xmm"#c1"; vfnmadd231ps %%xmm0,%%xmm2,%%xmm"#c2";"
+
+#define SOLVE_col2_ltor_m1n4(b_off,c1,...) SOLVE_col2_mul_m1n4(b_off,c1,__VA_ARGS__)\
+  "vblendps $3,%9,%%xmm0,%%xmm0; vfnmadd231ps %%xmm0,%%xmm1,%%xmm"#c1";"
+
+#define SOLVE_col2_ltor_m2n4(b_off,c1,c2,...) SOLVE_col2_mul_m2n4(b_off,c1,c2,__VA_ARGS__)\
+  "vblendps $3,%9,%%xmm0,%%xmm0; vfnmadd231ps %%xmm0,%%xmm1,%%xmm"#c1"; vfnmadd231ps %%xmm0,%%xmm2,%%xmm"#c2";"
+
+#define SOLVE_col3_mul_m1n4(b_off,c1,...)\
+  "vmovups "#b_off"("#__VA_ARGS__"),%%xmm0; vblendps $11,%8,%%xmm0,%%xmm2;"\
+  "vmulps %%xmm2,%%xmm"#c1",%%xmm"#c1";"\
+  "vpermilps $170,%%xmm"#c1",%%xmm1;"
+
+#define SOLVE_col3_mul_m2n4(b_off,c1,c2,...)\
+  "vmovups "#b_off"("#__VA_ARGS__"),%%xmm0; vblendps $11,%8,%%xmm0,%%xmm2;"\
+  "vmulps %%xmm2,%%xmm"#c1",%%xmm"#c1"; vmulps %%xmm2,%%xmm"#c2",%%xmm"#c2";"\
+  "vpermilps $170,%%xmm"#c1",%%xmm1; vpermilps $170,%%xmm"#c2",%%xmm2;"
+
+#define SOLVE_col3_rtol_m1n4(b_off,c1,...) SOLVE_col3_mul_m1n4(b_off,c1,__VA_ARGS__)\
+  "vblendps $12,%9,%%xmm0,%%xmm0; vfnmadd231ps %%xmm0,%%xmm1,%%xmm"#c1";"
+
+#define SOLVE_col3_rtol_m2n4(b_off,c1,c2,...) SOLVE_col3_mul_m2n4(b_off,c1,c2,__VA_ARGS__)\
+  "vblendps $12,%9,%%xmm0,%%xmm0; vfnmadd231ps %%xmm0,%%xmm1,%%xmm"#c1"; vfnmadd231ps %%xmm0,%%xmm2,%%xmm"#c2";"
+
+#define SOLVE_col3_ltor_m1n4(b_off,c1,...) SOLVE_col3_mul_m1n4(b_off,c1,__VA_ARGS__)\
+  "vblendps $7,%9,%%xmm0,%%xmm0; vfnmadd231ps %%xmm0,%%xmm1,%%xmm"#c1";"
+
+#define SOLVE_col3_ltor_m2n4(b_off,c1,c2,...) SOLVE_col3_mul_m2n4(b_off,c1,c2,__VA_ARGS__)\
+  "vblendps $7,%9,%%xmm0,%%xmm0; vfnmadd231ps %%xmm0,%%xmm1,%%xmm"#c1"; vfnmadd231ps %%xmm0,%%xmm2,%%xmm"#c2";"
+
+#define SOLVE_col4_ltor_m1n4(b_off,c1,...)\
+  "vmovups "#b_off"("#__VA_ARGS__"),%%xmm0; vblendps $7,%8,%%xmm0,%%xmm2;"\
+  "vmulps %%xmm2,%%xmm"#c1",%%xmm"#c1";"\
+  "vpermilps $255,%%xmm"#c1",%%xmm1;"
+
+#define SOLVE_col4_ltor_m2n4(b_off,c1,c2,...)\
+  "vmovups "#b_off"("#__VA_ARGS__"),%%xmm0; vblendps $7,%8,%%xmm0,%%xmm2;"\
+  "vmulps %%xmm2,%%xmm"#c1",%%xmm"#c1"; vmulps %%xmm2,%%xmm"#c2",%%xmm"#c2";"\
+  "vpermilps $255,%%xmm"#c1",%%xmm1; vpermilps $255,%%xmm"#c2",%%xmm2;"
+
+#define SOLVE_col4_rtol_m1n4(b_off,c1,...) SOLVE_col4_ltor_m1n4(b_off,c1,__VA_ARGS__)\
+  "vblendps $8,%9,%%xmm0,%%xmm0; vfnmadd231ps %%xmm0,%%xmm1,%%xmm"#c1";"
+
+#define SOLVE_col4_rtol_m2n4(b_off,c1,c2,...) SOLVE_col4_ltor_m2n4(b_off,c1,c2,__VA_ARGS__)\
+  "vblendps $8,%9,%%xmm0,%%xmm0; vfnmadd231ps %%xmm0,%%xmm1,%%xmm"#c1"; vfnmadd231ps %%xmm0,%%xmm2,%%xmm"#c2";"
+
+#define SUBTRACT_m4n2(b_off,c1,...) "vbroadcastsd "#b_off"("#__VA_ARGS__"),%%ymm0; vfnmadd231ps %%ymm0,%%ymm1,%%ymm"#c1";"
+
+#define SUBTRACT_m8n2(b_off,c1,c2,...) SUBTRACT_m4n2(b_off,c1,__VA_ARGS__) "vfnmadd231ps %%ymm0,%%ymm2,%%ymm"#c2";"
+
+#define SUBTRACT_m1n4(b_off,c1,...) "vmovups "#b_off"("#__VA_ARGS__"),%%xmm0; vfnmadd231ps %%xmm0,%%xmm1,%%xmm"#c1";"
+
+#define SUBTRACT_m2n4(b_off,c1,c2,...) SUBTRACT_m1n4(b_off,c1,__VA_ARGS__) "vfnmadd231ps %%xmm0,%%xmm2,%%xmm"#c2";"
+
+#define SAVE_SOLUTION_m8n2(c1,c2,a_off)\
+  "vunpcklps %%ymm"#c2",%%ymm"#c1",%%ymm0; vunpckhps %%ymm"#c2",%%ymm"#c1",%%ymm1;"\
+  "vunpcklpd %%ymm1,%%ymm0,%%ymm"#c1"; vunpckhpd %%ymm1,%%ymm0,%%ymm"#c2";"\
+  "vmovups %%ymm"#c1","#a_off"(%0); vmovups %%ymm"#c2","#a_off"+32(%0);"\
+  "vmovups %%ymm"#c1",(%3); vmovups %%ymm"#c2",(%3,%4,1); leaq (%3,%4,2),%3;"
+
+#define SAVE_SOLUTION_m4n2(c1,a_off)\
+  "vpermilps $216,%%ymm"#c1",%%ymm"#c1"; vpermpd $216,%%ymm"#c1",%%ymm"#c1";"\
+  "vmovups %%ymm"#c1","#a_off"(%0); vmovups %%xmm"#c1",(%3); vextractf128 $1,%%ymm"#c1",(%3,%4,1); leaq (%3,%4,2),%3;"
+
+#define SAVE_SOLUTION_m2n4(c1,c2,a_off)\
+  "vunpcklps %%xmm"#c2",%%xmm"#c1",%%xmm0; vmovups %%xmm0,"#a_off"(%0); vmovsd %%xmm0,(%3); vmovhpd %%xmm0,(%3,%4,1); leaq (%3,%4,2),%3;"\
+  "vunpckhps %%xmm"#c2",%%xmm"#c1",%%xmm0; vmovups %%xmm0,"#a_off"+16(%0); vmovsd %%xmm0,(%3); vmovhpd %%xmm0,(%3,%4,1); leaq (%3,%4,2),%3;"
+
+#define SAVE_SOLUTION_m1n4(c1,a_off)\
+  "vmovups %%xmm"#c1","#a_off"(%0); vmovss %%xmm"#c1",(%3); vextractps $1,%%xmm"#c1",(%3,%4,1); leaq (%3,%4,2),%3;"\
+  "vextractps $2,%%xmm"#c1",(%3); vextractps $3,%%xmm"#c1",(%3,%4,1); leaq (%3,%4,2),%3;"
diff --git a/kernel/zarch/KERNEL.Z14 b/kernel/zarch/KERNEL.Z14
index f6e3bec23..3510938a7 100644
--- a/kernel/zarch/KERNEL.Z14
+++ b/kernel/zarch/KERNEL.Z14
@@ -86,24 +86,24 @@ DGEMVTKERNEL = dgemv_t_4.c
 CGEMVTKERNEL = cgemv_t_4.c
 ZGEMVTKERNEL = zgemv_t_4.c
 
-STRMMKERNEL	= strmm8x4V.S
-DTRMMKERNEL	= trmm8x4V.S
+STRMMKERNEL	= gemm_vec.c
+DTRMMKERNEL	= gemm_vec.c
 CTRMMKERNEL	= ctrmm4x4V.S
 ZTRMMKERNEL	= ztrmm4x4V.S
 
-SGEMMKERNEL    =  strmm8x4V.S
-SGEMMINCOPY    = ../generic/gemm_ncopy_8.c
-SGEMMITCOPY    = ../generic/gemm_tcopy_8.c
-SGEMMONCOPY    = ../generic/gemm_ncopy_4.c
-SGEMMOTCOPY    = ../generic/gemm_tcopy_4.c
+SGEMMKERNEL    = gemm_vec.c
+ifneq ($(SGEMM_UNROLL_M),$(SGEMM_UNROLL_N))
+SGEMMINCOPY    = ../generic/gemm_ncopy_$(SGEMM_UNROLL_M).c
+SGEMMITCOPY    = ../generic/gemm_tcopy_$(SGEMM_UNROLL_M).c
 SGEMMINCOPYOBJ = sgemm_incopy$(TSUFFIX).$(SUFFIX)
 SGEMMITCOPYOBJ = sgemm_itcopy$(TSUFFIX).$(SUFFIX)
+endif
+SGEMMONCOPY    = ../generic/gemm_ncopy_$(SGEMM_UNROLL_N).c
+SGEMMOTCOPY    = ../generic/gemm_tcopy_$(SGEMM_UNROLL_N).c
 SGEMMONCOPYOBJ = sgemm_oncopy$(TSUFFIX).$(SUFFIX)
 SGEMMOTCOPYOBJ = sgemm_otcopy$(TSUFFIX).$(SUFFIX)
 
-
- 
-DGEMMKERNEL    =  gemm8x4V.S
+DGEMMKERNEL    = gemm_vec.c
 DGEMMINCOPY    = ../generic/gemm_ncopy_8.c
 DGEMMITCOPY    = ../generic/gemm_tcopy_8.c
 DGEMMONCOPY    = ../generic/gemm_ncopy_4.c
@@ -145,7 +145,3 @@ ZTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
 ZTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
 ZTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
 
-
-
-
-
diff --git a/kernel/zarch/gemm_vec.c b/kernel/zarch/gemm_vec.c
new file mode 100644
index 000000000..eb6d7700b
--- /dev/null
+++ b/kernel/zarch/gemm_vec.c
@@ -0,0 +1,475 @@
+/*
+ * Copyright (c) IBM Corporation 2020.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ *
+ *    1. Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *
+ *    2. Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in
+ *       the documentation and/or other materials provided with the
+ *       distribution.
+ *    3. Neither the name of the OpenBLAS project nor the names of
+ *       its contributors may be used to endorse or promote products
+ *       derived from this software without specific prior written
+ *       permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
+ * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+ * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+ * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+ * USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+#include "common.h"
+#include <vecintrin.h>
+
+#include <stdbool.h>
+#include <stdio.h>
+#include <stdlib.h>
+
+#ifdef COMPLEX
+#error "Handling for complex numbers is not supported in this kernel"
+#endif
+
+#ifdef DOUBLE
+#define UNROLL_M DGEMM_DEFAULT_UNROLL_M
+#define UNROLL_N DGEMM_DEFAULT_UNROLL_N
+#else
+#define UNROLL_M SGEMM_DEFAULT_UNROLL_M
+#define UNROLL_N SGEMM_DEFAULT_UNROLL_N
+#endif
+
+static const size_t unroll_m = UNROLL_M;
+static const size_t unroll_n = UNROLL_N;
+
+/* Handling of triangular matrices */
+#ifdef TRMMKERNEL
+static const bool trmm = true;
+static const bool left =
+#ifdef LEFT
+	true;
+#else
+	false;
+#endif
+
+static const bool backwards =
+#if defined(LEFT) != defined(TRANSA)
+	true;
+#else
+	false;
+#endif
+
+#else
+static const bool trmm = false;
+static const bool left = false;
+static const bool backwards = false;
+#endif /* TRMMKERNEL */
+
+/*
+ * Background:
+ *
+ * The algorithm of GotoBLAS / OpenBLAS breaks down the matrix multiplication
+ * problem by splitting all matrices into partitions multiple times, so that the
+ * submatrices fit into the L1 or L2 caches. As a result, each multiplication of
+ * submatrices can stream data fast from L1 and L2 caches. Inbetween, it copies
+ * and rearranges the submatrices to enable contiguous memory accesses to
+ * improve locality in both caches and TLBs.
+ *
+ * At the heart of the algorithm is this kernel, which multiplies, a "Block
+ * matrix" A (small dimensions) with a "Panel matrix" B (number of rows is
+ * small) and adds the result into a "Panel matrix" C; GotoBLAS calls this
+ * operation GEBP. This kernel further partitions GEBP twice, such that (1)
+ * submatrices of C and B fit into the L1 caches (GEBP_column_block) and (2) a
+ * block of C fits into the registers, while multiplying panels from A and B
+ * streamed from the L2 and L1 cache, respectively (GEBP_block).
+ *
+ *
+ * Algorithm GEBP(A, B, C, m, n, k, alpha):
+ *
+ * The problem is calculating C += alpha * (A * B)
+ * C is an m x n matrix, A is an m x k matrix, B is an k x n matrix.
+ *
+ * - C is in column-major-order, with an offset of ldc to the element in the
+ *   next column (same row).
+ * - A is in row-major-order yet stores SGEMM_UNROLL_M elements of each column
+ *   contiguously while walking along rows.
+ * - B is in column-major-order but packs SGEMM_UNROLL_N elements of a row
+ *   contiguously.
+ * If the numbers of rows and columns are not multiples of SGEMM_UNROLL_M or
+ * SGEMM_UNROLL_N, the remaining elements are arranged in blocks with power-of-2
+ * dimensions (e.g., 5 remaining columns would be in a block-of-4 and a
+ * block-of-1).
+ *
+ * Note that packing A and B into that form is taken care of by the caller in
+ * driver/level3/level3.c (actually done by "copy kernels").
+ *
+ * Steps:
+ * - Partition C and B into blocks of n_r (SGEMM_UNROLL_N) columns, C_j and B_j.
+ *   Now, B_j should fit into the L1 cache.
+ * - For each partition, calculate C_j += alpha * (A * B_j) by
+ *     (1) Calculate C_aux := A * B_j (see below)
+ *     (2) unpack C_j = C_j + alpha * C_aux
+ *
+ *
+ * Algorithm for Calculating C_aux:
+ *
+ * - Further partition C_aux and A into groups of m_r (SGEMM_UNROLL_M) rows,
+ *   such that the m_r x n_r-submatrix of C_aux can be held in registers. Each
+ *   submatrix of C_aux can be calculated independently, and the registers are
+ *   added back into C_j.
+ *
+ * - For each row-block of C_aux:
+ *   (uses a row block of A and full B_j)
+ *    - stream over all columns of A, multiply with elements from B and
+ *      accumulate in registers. (use different inner-kernels to exploit
+ *      vectorization for varying block sizes)
+ *    - add alpha * row block of C_aux back into C_j.
+ *
+ * Note that there are additional mechanics for handling triangular matrices,
+ * calculating B := alpha (A * B) where either of the matrices A or B can be
+ * triangular. In case of A, the macro "LEFT" is defined. In addition, A can
+ * optionally be transposed.
+ * The code effectively skips an "offset" number of columns in A and rows of B
+ * in each block, to save unnecessary work by exploiting the triangular nature.
+ * To handle all cases, the code discerns (1) a "left" mode when A is triangular
+ * and (2) "forward" / "backwards" modes where only the first "offset"
+ * columns/rows of A/B are used or where the first "offset" columns/rows are
+ * skipped, respectively.
+ *
+ * Reference:
+ *
+ * The summary above is based on staring at various kernel implementations and:
+ * K. Goto and R. A. Van de Geijn, Anatomy of High-Performance Matrix
+ * Multiplication, in ACM Transactions of Mathematical Software, Vol.  34, No.
+ * 3, May 2008.
+ */
+
+#define VLEN_BYTES 16
+#define VLEN_FLOATS (VLEN_BYTES / sizeof(FLOAT))
+
+typedef FLOAT vector_float __attribute__ ((vector_size (16)));
+
+/**
+ * Load a vector into register, and hint on 8-byte alignment to improve
+ * performance. gcc-9 and newer will create these hints by itself. For older
+ * compiler versions, use inline assembly to explicitly express the hint.
+ * Provide explicit hex encoding to cater for binutils versions that do not know
+ * about vector-load with alignment hints yet.
+ *
+ * Note that, for block sizes where we apply vectorization, vectors in A will
+ * always be 8-byte aligned.
+ */
+static inline vector_float vec_load_hinted(FLOAT const *restrict a) {
+	vector_float const *restrict addr = (vector_float const *restrict)a;
+	vector_float y;
+
+#if __GNUC__ < 9
+	// hex-encode vl %[out],%[addr],3
+	asm(".insn vrx,0xe70000003006,%[out],%[addr],3"
+	    : [ out ] "=v"(y)
+	    : [ addr ] "R"(*addr));
+#else
+	y = *addr;
+#endif
+
+	return y;
+}
+
+/**
+ * Calculate for a row-block in C_i of size ROWSxCOLS using vector intrinsics.
+ *
+ * @param[in] 	A	Pointer current block of input matrix A.
+ * @param[in]	k	Number of columns in A.
+ * @param[in]	B	Pointer current block of input matrix B.
+ * @param[inout] C	Pointer current block of output matrix C.
+ * @param[in]	ldc	Offset between elements in adjacent columns in C.
+ * @param[in]	alpha	Scalar factor.
+ */
+#define VECTOR_BLOCK(ROWS, COLS)                                              \
+	static inline void GEBP_block_##ROWS##_##COLS(                        \
+	    FLOAT const *restrict A, BLASLONG bk, FLOAT const *restrict B,    \
+	    FLOAT *restrict C, BLASLONG ldc, FLOAT alpha) {                   \
+		_Static_assert(                                               \
+		    ROWS % VLEN_FLOATS == 0,                                  \
+		    "rows in block must be multiples of vector length");      \
+		vector_float Caux[ROWS / VLEN_FLOATS][COLS];                  \
+                                                                              \
+		for (BLASLONG i = 0; i < ROWS / VLEN_FLOATS; i++) {           \
+			vector_float A0 =                                     \
+			    vec_load_hinted(A + i * VLEN_FLOATS);             \
+			for (BLASLONG j = 0; j < COLS; j++)                   \
+				Caux[i][j] = A0 * B[j];                       \
+		}                                                             \
+                                                                              \
+		/*                                                            \
+		 * Stream over the row-block of A, which is packed            \
+		 * column-by-column, multiply by coefficients in B and add up \
+		 * into temporaries Caux (which the compiler will hold in     \
+		 * registers). Vectorization: Multiply column vectors from A  \
+		 * with scalars from B and add up in column vectors of Caux.  \
+		 * That equates to unrolling the loop over rows (in i) and    \
+		 * executing each unrolled iteration as a vector element.     \
+		 */                                                           \
+		for (BLASLONG k = 1; k < bk; k++) {                           \
+			for (BLASLONG i = 0; i < ROWS / VLEN_FLOATS; i++) {   \
+				vector_float Ak = vec_load_hinted(            \
+				    A + i * VLEN_FLOATS + k * ROWS);          \
+                                                                              \
+				for (BLASLONG j = 0; j < COLS; j++)           \
+					Caux[i][j] += Ak * B[j + k * COLS];   \
+			}                                                     \
+		}                                                             \
+                                                                              \
+		/*                                                            \
+		 * Unpack row-block of C_aux into outer C_i, multiply by      \
+		 * alpha and add up.                                          \
+		 */                                                           \
+		for (BLASLONG j = 0; j < COLS; j++) {                         \
+			for (BLASLONG i = 0; i < ROWS / VLEN_FLOATS; i++) {   \
+				vector_float *C_ij =                          \
+				    (vector_float *)(C + i * VLEN_FLOATS +    \
+						     j * ldc);                \
+				if (trmm) {                                   \
+					*C_ij = alpha * Caux[i][j];           \
+				} else {                                      \
+					*C_ij += alpha * Caux[i][j];          \
+				}                                             \
+			}                                                     \
+		}                                                             \
+	}
+
+
+#if UNROLL_M == 16
+VECTOR_BLOCK(16, 4)
+VECTOR_BLOCK(16, 2)
+VECTOR_BLOCK(16, 1)
+#endif
+#if UNROLL_N == 8
+VECTOR_BLOCK(8, 8)
+VECTOR_BLOCK(4, 8)
+#endif
+VECTOR_BLOCK(8, 4)
+VECTOR_BLOCK(8, 2)
+VECTOR_BLOCK(8, 1)
+VECTOR_BLOCK(4, 4)
+VECTOR_BLOCK(4, 2)
+VECTOR_BLOCK(4, 1)
+
+#ifdef DOUBLE
+VECTOR_BLOCK(2, 4)
+VECTOR_BLOCK(2, 2)
+#endif
+
+/**
+ * Handle calculation for row blocks in C_i of any size by dispatching into
+ * macro-defined (inline) functions or by deferring to a simple generic
+ * implementation. Note that the compiler can remove this awkward-looking
+ * dispatching code while inlineing.
+ *
+ * @param[in]	m	Number of rows in block C_i.
+ * @param[in]	n	Number of columns in block C_i.
+ * @param[in]	first_row Index of first row of the block C_i (relative to C).
+ * @param[in]	A	Pointer to input matrix A (note: all of it).
+ * @param[in]	k	Number of columns in A and rows in B.
+ * @param[in]	B	Pointer to current column block (panel) of input matrix B.
+ * @param[inout] C	Pointer to current column block (panel) of output matrix C.
+ * @param[in]	ldc	Offset between elements in adjacent columns in C.
+ * @param[in]	alpha	Scalar factor.
+ * @param[in]	offset  Number of columns of A and rows of B to skip (for triangular matrices).
+ * @param[in]	off	Running offset for handling triangular matrices.
+ */
+static inline void GEBP_block(BLASLONG m, BLASLONG n,
+		       BLASLONG first_row,
+		       const FLOAT * restrict A, BLASLONG k,
+		       const FLOAT * restrict B,
+		       FLOAT *restrict C, BLASLONG ldc,
+		       FLOAT alpha,
+		       BLASLONG offset, BLASLONG off)
+{
+	if (trmm && left)
+		off = offset + first_row;
+
+	A += first_row * k;
+	C += first_row;
+
+	if (trmm) {
+		if (backwards) {
+			A += off * m;
+			B += off * n;
+			k -= off;
+		} else {
+			if (left) {
+				k = off + m;
+			} else {
+				k = off + n;
+			}
+		}
+	}
+
+#define BLOCK(bm, bn)                                           \
+	if (m == bm && n == bn) {                               \
+		GEBP_block_##bm##_##bn(A, k, B, C, ldc, alpha); \
+		return;                                         \
+	}
+
+#if UNROLL_M == 16
+	BLOCK(16, 4); BLOCK(16, 2); BLOCK(16, 1);
+#endif
+#if UNROLL_N == 8
+	BLOCK(8, 8); BLOCK(4, 8);
+#endif
+	BLOCK(8, 4); BLOCK(8, 2); BLOCK(8, 1);
+	BLOCK(4, 4); BLOCK(4, 2); BLOCK(4, 1);
+
+	#ifdef DOUBLE
+	BLOCK(2, 4);
+	BLOCK(2, 2);
+	#endif
+
+#undef BLOCK
+
+	/* simple implementation for smaller block sizes: */
+	FLOAT Caux[m][n] __attribute__ ((aligned (16)));
+
+	/*
+	 * Peel off first iteration (i.e., column of A) for initializing Caux
+	 */
+	for (BLASLONG i = 0; i < m; i++)
+		for (BLASLONG j = 0; j < n; j++)
+			Caux[i][j] = A[i] * B[j];
+
+	for (BLASLONG kk = 1; kk < k; kk++)
+		for (BLASLONG i = 0; i < m; i++)
+			for (BLASLONG j = 0; j < n; j++)
+				Caux[i][j] += A[i + kk * m] * B[j + kk * n];
+
+	for (BLASLONG i = 0; i < m; i++)
+		for (BLASLONG j = 0; j < n; j++)
+			if (trmm) {
+				C[i + j * ldc] = alpha * Caux[i][j];
+			} else {
+				C[i + j * ldc] += alpha * Caux[i][j];
+			}
+}
+
+/**
+ * Handle a column block (panel) of C and B while calculating C += alpha(A * B).
+ *
+ * @param[in]	num_cols	Number of columns in the block (in C and B).
+ * @param[in]	first_col	First column of the current block (in C and B).
+ * @param[in]	A	Pointer to input matrix A.
+ * @param[in]	bk	Number of columns in A and rows in B.
+ * @param[in]	B	Pointer to input matrix B (note: all of it).
+ * @param[in]	bm	Number of rows in C and A.
+ * @param[inout] C	Pointer to output matrix C (note: all of it).
+ * @param[in]	ldc	Offset between elements in adjacent columns in C.
+ * @param[in]	alpha	Scalar factor.
+ * @param[in]	offset	Number of columns of A and rows of B to skip (for triangular matrices).
+ */
+static inline void GEBP_column_block(BLASLONG num_cols, BLASLONG first_col,
+			const FLOAT *restrict A, BLASLONG bk,
+			const FLOAT *restrict B, BLASLONG bm,
+			FLOAT *restrict C, BLASLONG ldc,
+			FLOAT alpha,
+			BLASLONG const offset) {
+
+	FLOAT *restrict C_i = C + first_col * ldc;
+	/*
+	 * B is in column-order with n_r packed row elements, which does
+	 * not matter -- we always move in full such blocks of
+	 * column*pack
+	 */
+	const FLOAT *restrict B_i = B + first_col * bk;
+
+	BLASLONG off = 0;
+	if (trmm) {
+		if (left) {
+			off = offset;
+		} else {
+			off = -offset + first_col;
+		}
+	}
+
+	/*
+	 * Calculate C_aux := A * B_j
+	 * then unpack C_i += alpha * C_aux.
+	 *
+	 * For that purpose, further partition C_aux and A into blocks
+	 * of m_r (unroll_m) rows, or powers-of-2 if smaller.
+	 */
+	BLASLONG row = 0;
+	for (BLASLONG block_size = unroll_m; block_size > 0; block_size /= 2)
+		for (; bm - row >= block_size; row += block_size)
+			GEBP_block(block_size, num_cols, row, A, bk, B_i, C_i,
+				   ldc, alpha, offset, off);
+}
+
+/**
+ * Inner kernel for matrix-matrix multiplication. C += alpha (A * B)
+ * where C is an m-by-n matrix, A is m-by-k and B is k-by-n. Note that A, B, and
+ * C are pointers to submatrices of the actual matrices.
+ *
+ * For triangular matrix multiplication, calculate B := alpha (A * B) where A
+ * or B can be triangular (in case of A, the macro LEFT will be defined).
+ *
+ * @param[in]	bm	Number of rows in C and A.
+ * @param[in]	bn	Number of columns in C and B.
+ * @param[in]	bk	Number of columns in A and rows in B.
+ * @param[in]	alpha	Scalar factor.
+ * @param[in]	ba	Pointer to input matrix A.
+ * @param[in]	bb	Pointer to input matrix B.
+ * @param[inout] C	Pointer to output matrix C.
+ * @param[in]	ldc	Offset between elements in adjacent columns in C.
+ * @param[in]	offset	Number of columns of A and rows of B to skip (for triangular matrices).
+ * @returns 0 on success.
+ */
+int CNAME(BLASLONG bm, BLASLONG bn, BLASLONG bk, FLOAT alpha,
+	  FLOAT *restrict ba, FLOAT *restrict bb,
+	  FLOAT *restrict C, BLASLONG ldc
+#ifdef TRMMKERNEL
+	  , BLASLONG offset
+#endif
+	  )
+{
+	if ( (bm == 0) || (bn == 0) || (bk == 0) || (alpha == ZERO))
+		return 0;
+
+	/*
+	 * interface code allocates buffers for ba and bb at page
+	 * granularity (i.e., using mmap(MAP_ANONYMOUS), so enable the compiler
+	 * to make use of the fact in vector load operations.
+	 */
+	ba = __builtin_assume_aligned(ba, 16);
+	bb = __builtin_assume_aligned(bb, 16);
+
+	/*
+	 * Use offset and off even when compiled as SGEMMKERNEL to simplify
+	 * function signatures and function calls.
+	 */
+#ifndef TRMMKERNEL
+	BLASLONG const offset = 0;
+#endif
+
+	/*
+	 * Partition B and C into blocks of n_r (unroll_n) columns, called B_i
+	 * and C_i. For each partition, calculate C_i += alpha * (A * B_j).
+	 *
+	 * For remaining columns that do not fill up a block of n_r, iteratively
+	 * use smaller block sizes of powers of 2.
+	 */
+	BLASLONG col = 0;
+	for (BLASLONG block_size = unroll_n; block_size > 0; block_size /= 2)
+		for (; bn - col >= block_size; col += block_size)
+			GEBP_column_block(block_size, col, ba, bk, bb, bm, C, ldc, alpha, offset);
+
+   return 0;
+}
diff --git a/lapack-netlib/LAPACKE/include/lapack.h b/lapack-netlib/LAPACKE/include/lapack.h
index 0a6226fe4..36e53ec24 100644
--- a/lapack-netlib/LAPACKE/include/lapack.h
+++ b/lapack-netlib/LAPACKE/include/lapack.h
@@ -12,27 +12,6 @@
 
 #include <stdlib.h>
 
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-/*----------------------------------------------------------------------------*/
-#ifndef lapack_int
-#define lapack_int     int
-#endif
-
-#ifndef lapack_logical
-#define lapack_logical lapack_int
-#endif
-
-/* f2c, hence clapack and MacOS Accelerate, returns double instead of float
- * for sdot, slange, clange, etc. */
-#if defined(LAPACK_F2C)
-    typedef double lapack_float_return;
-#else
-    typedef float lapack_float_return;
-#endif
-
 /* Complex types are structures equivalent to the
 * Fortran complex types COMPLEX(4) and COMPLEX(8).
 *
@@ -88,6 +67,29 @@ extern "C" {
 
 #endif /* LAPACK_COMPLEX_CUSTOM */
 
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/*----------------------------------------------------------------------------*/
+#ifndef lapack_int
+#define lapack_int     int
+#endif
+
+#ifndef lapack_logical
+#define lapack_logical lapack_int
+#endif
+
+/* f2c, hence clapack and MacOS Accelerate, returns double instead of float
+ * for sdot, slange, clange, etc. */
+#if defined(LAPACK_F2C)
+    typedef double lapack_float_return;
+#else
+    typedef float lapack_float_return;
+#endif
+
+
 /* Callback logical functions of one, two, or three arguments are used
 *  to select eigenvalues to sort to the top left of the Schur form.
 *  The value is selected if function returns TRUE (non-zero). */
diff --git a/lapack-netlib/LAPACKE/include/lapacke.h b/lapack-netlib/LAPACKE/include/lapacke.h
index 6eb0b696b..012c104bb 100644
--- a/lapack-netlib/LAPACKE/include/lapacke.h
+++ b/lapack-netlib/LAPACKE/include/lapacke.h
@@ -12575,7 +12575,7 @@ lapack_int LAPACKE_zhetrs_aa_2stage_work( int matrix_layout, char uplo, lapack_i
                    
 /* APIs for set/get nancheck flags */
 void LAPACKE_set_nancheck( int flag );
-int LAPACKE_get_nancheck( );
+int LAPACKE_get_nancheck( void );
 
 #ifdef __cplusplus
 }
diff --git a/lapack-netlib/SRC/clargv.f b/lapack-netlib/SRC/clargv.f
index ba53cae6f..36c5108df 100644
--- a/lapack-netlib/SRC/clargv.f
+++ b/lapack-netlib/SRC/clargv.f
@@ -200,7 +200,7 @@
             FS = FS*SAFMN2
             GS = GS*SAFMN2
             SCALE = SCALE*SAFMN2
-            IF( SCALE.GE.SAFMX2 )
+            IF( SCALE.GE.SAFMX2 .AND. COUNT .LT. 20 )
      $         GO TO 10
          ELSE IF( SCALE.LE.SAFMN2 ) THEN
             IF( G.EQ.CZERO ) THEN
diff --git a/lapack-netlib/SRC/clartg.f b/lapack-netlib/SRC/clartg.f
index da9a1cdef..baa68b657 100644
--- a/lapack-netlib/SRC/clartg.f
+++ b/lapack-netlib/SRC/clartg.f
@@ -161,7 +161,7 @@
          FS = FS*SAFMN2
          GS = GS*SAFMN2
          SCALE = SCALE*SAFMN2
-         IF( SCALE.GE.SAFMX2 )
+         IF( SCALE.GE.SAFMX2 .AND. COUNT .LT. 20)
      $      GO TO 10
       ELSE IF( SCALE.LE.SAFMN2 ) THEN
          IF( G.EQ.CZERO.OR.SISNAN( ABS( G ) ) ) THEN
diff --git a/lapack-netlib/SRC/dlartg.f b/lapack-netlib/SRC/dlartg.f
index 1c7c46f63..dc49986a0 100644
--- a/lapack-netlib/SRC/dlartg.f
+++ b/lapack-netlib/SRC/dlartg.f
@@ -163,7 +163,7 @@
             F1 = F1*SAFMN2
             G1 = G1*SAFMN2
             SCALE = MAX( ABS( F1 ), ABS( G1 ) )
-            IF( SCALE.GE.SAFMX2 )
+            IF( SCALE.GE.SAFMX2 .AND. COUNT .LT. 20)
      $         GO TO 10
             R = SQRT( F1**2+G1**2 )
             CS = F1 / R
diff --git a/lapack-netlib/SRC/dlartgp.f b/lapack-netlib/SRC/dlartgp.f
index 0cb0d2d13..334e416e8 100644
--- a/lapack-netlib/SRC/dlartgp.f
+++ b/lapack-netlib/SRC/dlartgp.f
@@ -161,7 +161,7 @@
             F1 = F1*SAFMN2
             G1 = G1*SAFMN2
             SCALE = MAX( ABS( F1 ), ABS( G1 ) )
-            IF( SCALE.GE.SAFMX2 )
+            IF( SCALE.GE.SAFMX2 .AND. COUNT .LT. 20 )
      $         GO TO 10
             R = SQRT( F1**2+G1**2 )
             CS = F1 / R
diff --git a/lapack-netlib/SRC/slartg.f b/lapack-netlib/SRC/slartg.f
index 784d4bc36..307c9c83a 100644
--- a/lapack-netlib/SRC/slartg.f
+++ b/lapack-netlib/SRC/slartg.f
@@ -163,7 +163,7 @@
             F1 = F1*SAFMN2
             G1 = G1*SAFMN2
             SCALE = MAX( ABS( F1 ), ABS( G1 ) )
-            IF( SCALE.GE.SAFMX2 )
+            IF( SCALE.GE.SAFMX2 .AND. COUNT .LT. 20)
      $         GO TO 10
             R = SQRT( F1**2+G1**2 )
             CS = F1 / R
diff --git a/lapack-netlib/SRC/slartgp.f b/lapack-netlib/SRC/slartgp.f
index ad76c94b4..f8be5f52b 100644
--- a/lapack-netlib/SRC/slartgp.f
+++ b/lapack-netlib/SRC/slartgp.f
@@ -161,7 +161,7 @@
             F1 = F1*SAFMN2
             G1 = G1*SAFMN2
             SCALE = MAX( ABS( F1 ), ABS( G1 ) )
-            IF( SCALE.GE.SAFMX2 )
+            IF( SCALE.GE.SAFMX2 .AND. COUNT .LT. 20)
      $         GO TO 10
             R = SQRT( F1**2+G1**2 )
             CS = F1 / R
diff --git a/lapack-netlib/SRC/sorhr_col.f b/lapack-netlib/SRC/sorhr_col.f
index 38976245c..9aef57b26 100644
--- a/lapack-netlib/SRC/sorhr_col.f
+++ b/lapack-netlib/SRC/sorhr_col.f
@@ -282,7 +282,8 @@
      $                   NPLUSONE
 *     ..
 *     .. External Subroutines ..
-      EXTERNAL           SCOPY, SLAORHR_COL_GETRFNP, SSCAL, STRSM, XERBLA
+      EXTERNAL           SCOPY, SLAORHR_COL_GETRFNP, SSCAL, STRSM,
+     $XERBLA
 *     ..
 *     .. Intrinsic Functions ..
       INTRINSIC          MAX, MIN
@@ -436,4 +437,4 @@
 *
 *     End of SORHR_COL
 *
-      END
\ No newline at end of file
+      END
diff --git a/lapack-netlib/SRC/zheequb.f b/lapack-netlib/SRC/zheequb.f
index d698232e8..7d719f41e 100644
--- a/lapack-netlib/SRC/zheequb.f
+++ b/lapack-netlib/SRC/zheequb.f
@@ -271,7 +271,7 @@
          AVG = AVG / N
 
          STD = 0.0D0
-         DO I = N+1, N
+         DO I = N+1, 2*N
             WORK( I ) = S( I-N ) * WORK( I-N ) - AVG
          END DO
          CALL ZLASSQ( N, WORK( N+1 ), 1, SCALE, SUMSQ )
diff --git a/lapack-netlib/SRC/zlargv.f b/lapack-netlib/SRC/zlargv.f
index 1e17983d5..f83ca1851 100644
--- a/lapack-netlib/SRC/zlargv.f
+++ b/lapack-netlib/SRC/zlargv.f
@@ -201,7 +201,7 @@
             FS = FS*SAFMN2
             GS = GS*SAFMN2
             SCALE = SCALE*SAFMN2
-            IF( SCALE.GE.SAFMX2 )
+            IF( SCALE.GE.SAFMX2 .AND. COUNT .LT. 20 )
      $         GO TO 10
          ELSE IF( SCALE.LE.SAFMN2 ) THEN
             IF( G.EQ.CZERO ) THEN
diff --git a/lapack-netlib/SRC/zlartg.f b/lapack-netlib/SRC/zlartg.f
index 8989bb896..894b4ded0 100644
--- a/lapack-netlib/SRC/zlartg.f
+++ b/lapack-netlib/SRC/zlartg.f
@@ -161,7 +161,7 @@
          FS = FS*SAFMN2
          GS = GS*SAFMN2
          SCALE = SCALE*SAFMN2
-         IF( SCALE.GE.SAFMX2 )
+         IF( SCALE.GE.SAFMX2 .AND. COUNT .LT. 20 )
      $      GO TO 10
       ELSE IF( SCALE.LE.SAFMN2 ) THEN
          IF( G.EQ.CZERO.OR.DISNAN( ABS( G ) ) ) THEN
diff --git a/lapack/CMakeLists.txt b/lapack/CMakeLists.txt
index e21a9aabb..778e6f8fa 100644
--- a/lapack/CMakeLists.txt
+++ b/lapack/CMakeLists.txt
@@ -2,6 +2,7 @@
 include_directories(${PROJECT_SOURCE_DIR})
 include_directories(${PROJECT_BINARY_DIR})
 
+list (REMOVE_ITEM FLOAT_TYPES "HALF")
 
 set(LAPACK_SOURCES
   potrf/potrf_U_single.c
@@ -45,6 +46,9 @@ GenerateNamedObjects("laswp/generic/laswp_k_4.c" "" "laswp_plus" false "" ""  fa
 GenerateNamedObjects("laswp/generic/laswp_k_4.c" "MINUS" "laswp_minus" false "" ""  false 3)
 
 foreach (float_type ${FLOAT_TYPES})
+if (${float_type} STREQUAL "HALF")
+ continue()
+endif()
 GenerateNamedObjects("getrf/getrf_single.c" "UNIT" "getrf_single" false "" "" false ${float_type})
 endforeach ()
 
diff --git a/lapack/getrf/getrf_parallel.c b/lapack/getrf/getrf_parallel.c
index c82defcab..c602822a8 100644
--- a/lapack/getrf/getrf_parallel.c
+++ b/lapack/getrf/getrf_parallel.c
@@ -68,23 +68,14 @@ double sqrt(double);
 #define GETRF_FACTOR 1.00
 
 
-#if   defined(USE_PTHREAD_LOCK)
-static pthread_mutex_t    getrf_lock = PTHREAD_MUTEX_INITIALIZER;
-#elif defined(USE_PTHREAD_SPINLOCK)
-static pthread_spinlock_t getrf_lock = 0;
+#if (__STDC_VERSION__ >= 201112L)
+#define	atomic_load_long(p)		__atomic_load_n(p, __ATOMIC_RELAXED)
+#define	atomic_store_long(p, v)		__atomic_store_n(p, v, __ATOMIC_RELAXED)
 #else
-static BLASULONG  getrf_lock = 0UL;
+#define	atomic_load_long(p)		(BLASLONG)(*(volatile BLASLONG*)(p))
+#define	atomic_store_long(p, v)		(*(volatile BLASLONG *)(p)) = (v)
 #endif
 
-#if   defined(USE_PTHREAD_LOCK)
-static pthread_mutex_t    getrf_flag_lock = PTHREAD_MUTEX_INITIALIZER;
-#elif defined(USE_PTHREAD_SPINLOCK)
-static pthread_spinlock_t getrf_flag_lock = 0;
-#else
-static BLASULONG  getrf_flag_lock = 0UL;
-#endif
-
-
 
 
 static __inline BLASLONG FORMULA1(BLASLONG M, BLASLONG N, BLASLONG IS, BLASLONG BK, BLASLONG T) {
@@ -119,11 +110,7 @@ static void inner_basic_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *ra
   FLOAT *d = (FLOAT *)args -> b + (k + k * lda) * COMPSIZE;
   FLOAT *sbb = sb;
 
-#if __STDC_VERSION__ >= 201112L
-  _Atomic BLASLONG *flag = (_Atomic BLASLONG *)args -> d;
-#else
   volatile BLASLONG *flag = (volatile BLASLONG *)args -> d;
-#endif
 
   blasint *ipiv = (blasint *)args -> c;
 
@@ -180,7 +167,10 @@ static void inner_basic_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *ra
       }
     }
 
-    if ((js + REAL_GEMM_R >= n) && (mypos >= 0)) flag[mypos * CACHE_LINE_SIZE] = 0;
+    if ((js + REAL_GEMM_R >= n) && (mypos >= 0)) {
+	    MB;
+	    atomic_store_long(&flag[mypos * CACHE_LINE_SIZE], 0);
+    }
 
     for (is = 0; is < m; is += GEMM_P){
       min_i = m - is;
@@ -201,14 +191,10 @@ static void inner_basic_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *ra
 /* Non blocking implementation */
 
 typedef struct {
-#if __STDC_VERSION__ >= 201112L
-  _Atomic
-#else
-  volatile
-#endif
-   BLASLONG working[MAX_CPU_NUMBER][CACHE_LINE_SIZE * DIVIDE_RATE];
+  volatile BLASLONG working[MAX_CPU_NUMBER][CACHE_LINE_SIZE * DIVIDE_RATE];
 } job_t;
 
+
 #define ICOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_ITCOPY(M, N, (FLOAT *)(A) + ((Y) + (X) * (LDA)) * COMPSIZE, LDA, BUFFER);
 #define OCOPY_OPERATION(M, N, A, LDA, X, Y, BUFFER) GEMM_ONCOPY(M, N, (FLOAT *)(A) + ((X) + (Y) * (LDA)) * COMPSIZE, LDA, BUFFER);
 
@@ -246,11 +232,8 @@ static int inner_advanced_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *
 
   blasint *ipiv = (blasint *)args -> c;
   BLASLONG jw;
-#if __STDC_VERSION__ >= 201112L
-  _Atomic BLASLONG *flag = (_Atomic BLASLONG *)args -> d;
-#else
   volatile BLASLONG *flag = (volatile BLASLONG *)args -> d;
-#endif
+
   if (args -> a == NULL) {
     TRSM_ILTCOPY(k, k, (FLOAT *)args -> b, lda, 0, sb);
     sbb = (FLOAT *)((((BLASULONG)(sb + k * k * COMPSIZE) + GEMM_ALIGN) & ~GEMM_ALIGN) + GEMM_OFFSET_B);
@@ -280,10 +263,9 @@ static int inner_advanced_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *
 #if 1
     {
 	do {
-	    LOCK_COMMAND(&getrf_lock);
-	    jw = job[mypos].working[i][CACHE_LINE_SIZE * bufferside];
-	    UNLOCK_COMMAND(&getrf_lock);
+	   jw =  atomic_load_long(&job[mypos].working[i][CACHE_LINE_SIZE * bufferside]);
 	} while (jw);
+	MB;
     }
 #else
       while (job[mypos].working[i][CACHE_LINE_SIZE * bufferside]) {};
@@ -326,21 +308,17 @@ static int inner_advanced_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *
     }
     MB;
     for (i = 0; i < args -> nthreads; i++) {
-      LOCK_COMMAND(&getrf_lock);
-      job[mypos].working[i][CACHE_LINE_SIZE * bufferside] = (BLASLONG)buffer[bufferside];
-      UNLOCK_COMMAND(&getrf_lock);
+      atomic_store_long(&job[mypos].working[i][CACHE_LINE_SIZE * bufferside], (BLASLONG)buffer[bufferside]);
     }
   }
 
-  LOCK_COMMAND(&getrf_flag_lock);
-  flag[mypos * CACHE_LINE_SIZE] = 0;
-  UNLOCK_COMMAND(&getrf_flag_lock);
+  MB;
+  atomic_store_long(&flag[mypos * CACHE_LINE_SIZE], 0);
 
   if (m == 0) {
+    MB;
     for (xxx = 0; xxx < DIVIDE_RATE; xxx++) {
-      LOCK_COMMAND(&getrf_lock);
-      job[mypos].working[mypos][CACHE_LINE_SIZE * xxx] = 0;
-      UNLOCK_COMMAND(&getrf_lock);
+      atomic_store_long(&job[mypos].working[mypos][CACHE_LINE_SIZE * xxx], 0);
     }
   }
 
@@ -366,10 +344,9 @@ static int inner_advanced_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *
 	  if ((current != mypos) && (!is)) {
 #if 1
 		do {
-		    LOCK_COMMAND(&getrf_lock);
-		    jw = job[current].working[mypos][CACHE_LINE_SIZE * bufferside];
-		    UNLOCK_COMMAND(&getrf_lock);
-		} while (jw == 0);
+		   jw =  atomic_load_long(&job[current].working[mypos][CACHE_LINE_SIZE * bufferside]);
+	        } while (jw == 0);
+		MB;
 #else
 	    	    while(job[current].working[mypos][CACHE_LINE_SIZE * bufferside] == 0) {};
 #endif
@@ -381,9 +358,7 @@ static int inner_advanced_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *
 
 	  MB;
 	  if (is + min_i >= m) {
-            LOCK_COMMAND(&getrf_lock);
-	    job[current].working[mypos][CACHE_LINE_SIZE * bufferside] = 0;
-            UNLOCK_COMMAND(&getrf_lock);
+	    atomic_store_long(&job[current].working[mypos][CACHE_LINE_SIZE * bufferside], 0);
 	  }
 	}
 
@@ -397,10 +372,9 @@ static int inner_advanced_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *
     for (xxx = 0; xxx < DIVIDE_RATE; xxx++) {
 #if 1
 	do {
-	    LOCK_COMMAND(&getrf_lock);
-	    jw = job[mypos].working[i][CACHE_LINE_SIZE *xxx];
-	    UNLOCK_COMMAND(&getrf_lock);
+	    jw = atomic_load_long(&job[mypos].working[i][CACHE_LINE_SIZE *xxx]);
 	} while(jw != 0);
+	MB;
 #else
       while (job[mypos].working[i][CACHE_LINE_SIZE * xxx] ) {};
 #endif
@@ -443,12 +417,7 @@ blasint CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa,
 #ifdef _MSC_VER
   BLASLONG flag[MAX_CPU_NUMBER * CACHE_LINE_SIZE];
 #else
-#if __STDC_VERSION__ >= 201112L
-  _Atomic
-#else  
-  volatile
-#endif  
-   BLASLONG flag[MAX_CPU_NUMBER * CACHE_LINE_SIZE] __attribute__((aligned(128)));
+  volatile BLASLONG flag[MAX_CPU_NUMBER * CACHE_LINE_SIZE] __attribute__((aligned(128)));
 #endif
 
 #ifndef COMPLEX
@@ -543,7 +512,11 @@ blasint CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa,
       if (width > mn - is - bk) width = mn - is - bk;
     }
 
-    if (num_cpu > 0) exec_blas_async_wait(num_cpu, &queue[0]);
+
+    if (num_cpu > 0) {
+	    WMB;
+	    exec_blas_async_wait(num_cpu, &queue[0]);
+    }
 
     mm = m - bk - is;
     nn = n - bk - is;
@@ -608,7 +581,7 @@ blasint CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa,
       queue[num_cpu].sa      = NULL;
       queue[num_cpu].sb      = NULL;
       queue[num_cpu].next    = &queue[num_cpu + 1];
-      flag[num_cpu * CACHE_LINE_SIZE] = 1;
+      atomic_store_long(&flag[num_cpu * CACHE_LINE_SIZE], 1);
 
       num_cpu ++;
 
@@ -637,6 +610,8 @@ blasint CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa,
     if (num_cpu > 0) {
       queue[num_cpu - 1].next = NULL;
 
+      WMB;
+
       exec_blas_async(0, &queue[0]);
 
       inner_basic_thread(&newarg, NULL, range_n_mine, sa, sbb, -1);
@@ -647,14 +622,10 @@ blasint CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa,
 
       for (i = 0; i < num_cpu; i ++) {
 #if 1
-	      LOCK_COMMAND(&getrf_flag_lock);
-	      f=flag[i*CACHE_LINE_SIZE];
-	      UNLOCK_COMMAND(&getrf_flag_lock);
-	      while (f!=0) {
-	      LOCK_COMMAND(&getrf_flag_lock);
-	      f=flag[i*CACHE_LINE_SIZE];
-	      UNLOCK_COMMAND(&getrf_flag_lock);
-	      };
+	      do {
+		   f =  atomic_load_long(&flag[i*CACHE_LINE_SIZE]);
+	      } while (f != 0);
+	      MB;
 #else
               while (flag[i*CACHE_LINE_SIZE]) {};
 #endif
@@ -719,12 +690,7 @@ blasint CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa,
   BLASLONG range[MAX_CPU_NUMBER + 1];
 
   BLASLONG width, nn, num_cpu;
-#if __STDC_VERSION__ >= 201112L
-  _Atomic
-#else  
-  volatile
-#endif
-   BLASLONG flag[MAX_CPU_NUMBER * CACHE_LINE_SIZE] __attribute__((aligned(128)));
+  volatile BLASLONG flag[MAX_CPU_NUMBER * CACHE_LINE_SIZE] __attribute__((aligned(128)));
 
 #ifndef COMPLEX
 #ifdef XDOUBLE
@@ -833,6 +799,8 @@ blasint CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa,
     nn = n - bk - is;
     if (width > nn) width = nn;
 
+    WMB;
+
     if (num_cpu > 1)  exec_blas_async_wait(num_cpu - 1, &queue[1]);
 
     range[0] = 0;
@@ -867,7 +835,7 @@ blasint CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa,
       queue[num_cpu].sa      = NULL;
       queue[num_cpu].sb      = NULL;
       queue[num_cpu].next    = &queue[num_cpu + 1];
-      flag[num_cpu * CACHE_LINE_SIZE] = 1;
+      atomic_store_long(&flag[num_cpu * CACHE_LINE_SIZE], 1);
 
       num_cpu ++;
     }
@@ -882,6 +850,7 @@ blasint CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa,
     range_n_new[0] = offset + is;
     range_n_new[1] = offset + is + bk;
 
+    WMB;
     if (num_cpu > 1) {
 
       exec_blas_async(1, &queue[1]);
@@ -917,7 +886,7 @@ blasint CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa,
 
 #endif
 
-      for (i = 1; i < num_cpu; i ++) while (flag[i * CACHE_LINE_SIZE]) {};
+      for (i = 1; i < num_cpu; i ++) while (atomic_load_long(&flag[i * CACHE_LINE_SIZE])) {};
 
       TRSM_ILTCOPY(bk, bk, a + (is +  is * lda) * COMPSIZE, lda, 0, sb);
 
diff --git a/lapack/getrf/potrf_parallel.c b/lapack/getrf/potrf_parallel.c
index c2fee6bd1..312509685 100644
--- a/lapack/getrf/potrf_parallel.c
+++ b/lapack/getrf/potrf_parallel.c
@@ -380,6 +380,9 @@ static int thread_driver(blas_arg_t *args, FLOAT *sa, FLOAT *sb){
 #elif defined(DOUBLE)
   mode  =  BLAS_DOUBLE  | BLAS_REAL;
   mask  = MAX(DGEMM_UNROLL_M, DGEMM_UNROLL_N) - 1;
+#elif defined(HALF)
+  mode  =  BLAS_HALF  | BLAS_REAL;
+  mask  = MAX(SHGEMM_UNROLL_M, SHGEMM_UNROLL_N) - 1;
 #else
   mode  =  BLAS_SINGLE  | BLAS_REAL;
   mask  = MAX(SGEMM_UNROLL_M, SGEMM_UNROLL_N) - 1;
diff --git a/openblas_config_template.h b/openblas_config_template.h
index 52dd49da2..49aea1cab 100644
--- a/openblas_config_template.h
+++ b/openblas_config_template.h
@@ -91,3 +91,8 @@ typedef int blasint;
   #define openblas_complex_xdouble_real(z)           ((z).real)
   #define openblas_complex_xdouble_imag(z)           ((z).imag)
 #endif
+
+/* Inclusion of Linux-specific header is needed for definition of cpu_set_t. */
+#ifdef OPENBLAS_OS_LINUX
+#include <sched.h>
+#endif
diff --git a/param.h b/param.h
index 68e429734..04928277c 100644
--- a/param.h
+++ b/param.h
@@ -72,6 +72,12 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #ifndef PARAM_H
 #define PARAM_H
 
+#define SHGEMM_DEFAULT_UNROLL_N 4
+#define SHGEMM_DEFAULT_UNROLL_M 8
+#define SHGEMM_DEFAULT_UNROLL_MN 32
+#define SHGEMM_DEFAULT_P 256
+#define SHGEMM_DEFAULT_R 256
+#define SHGEMM_DEFAULT_Q 256
 #ifdef OPTERON
 
 #define SNUMOPT		4
@@ -1700,12 +1706,12 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #else
 
-#define SGEMM_DEFAULT_P 640
+#define SGEMM_DEFAULT_P 448
 #define DGEMM_DEFAULT_P 192
 #define CGEMM_DEFAULT_P 384
 #define ZGEMM_DEFAULT_P 256
 
-#define SGEMM_DEFAULT_Q 320
+#define SGEMM_DEFAULT_Q 448
 #define DGEMM_DEFAULT_Q 384
 #define CGEMM_DEFAULT_Q 192
 #define ZGEMM_DEFAULT_Q 128
@@ -1968,7 +1974,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define SGEMM_DEFAULT_UNROLL_N 4
 #define DGEMM_DEFAULT_UNROLL_M 4
 #define DGEMM_DEFAULT_UNROLL_N 4
-#define CGEMM_DEFAULT_UNROLL_M 8
+#define CGEMM_DEFAULT_UNROLL_M 2
 #define CGEMM_DEFAULT_UNROLL_N 2
 #define ZGEMM_DEFAULT_UNROLL_M 2
 #define ZGEMM_DEFAULT_UNROLL_N 2
@@ -2229,15 +2235,26 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define ZGEMM_DEFAULT_UNROLL_M 8
 #define ZGEMM_DEFAULT_UNROLL_N 2
 
-#define SGEMM_DEFAULT_P  1280
-#define DGEMM_DEFAULT_P  640
-#define CGEMM_DEFAULT_P  640
-#define ZGEMM_DEFAULT_P  320
+#define SGEMM_DEFAULT_P  1280UL
+#define DGEMM_DEFAULT_P  640UL
+#define CGEMM_DEFAULT_P  640UL
+#define ZGEMM_DEFAULT_P  320UL
 
-#define SGEMM_DEFAULT_Q  640
-#define DGEMM_DEFAULT_Q  720
-#define CGEMM_DEFAULT_Q  640
-#define ZGEMM_DEFAULT_Q  640
+#define SGEMM_DEFAULT_Q  640UL
+#define DGEMM_DEFAULT_Q  720UL
+#define CGEMM_DEFAULT_Q  640UL
+#define ZGEMM_DEFAULT_Q  640UL
+
+#if 0
+#define SGEMM_DEFAULT_R SGEMM_DEFAULT_P
+#define DGEMM_DEFAULT_R DGEMM_DEFAULT_P
+#define CGEMM_DEFAULT_R CGEMM_DEFAULT_P
+#define ZGEMM_DEFAULT_R ZGEMM_DEFAULT_P
+#endif
+#define SGEMM_DEFAULT_R 4096
+#define DGEMM_DEFAULT_R 4096
+#define CGEMM_DEFAULT_R 4096
+#define ZGEMM_DEFAULT_R 4096
 
 #define SYMV_P	 8
 
@@ -2271,6 +2288,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define CGEMM_DEFAULT_Q  1026
 #define ZGEMM_DEFAULT_Q 1026
 
+#define SGEMM_DEFAULT_R 4096
+#define DGEMM_DEFAULT_R 4096
+#define CGEMM_DEFAULT_R 4096
+#define ZGEMM_DEFAULT_R 4096
+
 #define SYMV_P	 8
 
 #endif
@@ -2457,7 +2479,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define SYMV_P	16
 #endif
 
-#if defined(P5600) || defined(MIPS1004K) || defined(I6400) || defined(P6600) || defined(I6500)
+#if defined(P5600) || defined(MIPS1004K) || defined(MIPS24K) || defined(I6400) || defined(P6600) || defined(I6500)
 #define SNUMOPT  2
 #define DNUMOPT  2
 
@@ -2601,7 +2623,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define SYMV_P	16
 
-#if defined(CORTEXA53) || defined(CORTEXA57) || \
+#if defined(CORTEXA57) || \
     defined(CORTEXA72) || defined(CORTEXA73) || \
     defined(FALKOR)    || defined(TSV110) || defined(EMAG8180)
 
@@ -2647,6 +2669,35 @@ is a big desktop or server with abundant cache rather than a phone or embedded d
 #define CGEMM_DEFAULT_R 4096
 #define ZGEMM_DEFAULT_R 2048
 
+#elif defined(CORTEXA53)
+
+#define SGEMM_DEFAULT_UNROLL_M  8
+#define SGEMM_DEFAULT_UNROLL_N  8
+
+#define DGEMM_DEFAULT_UNROLL_M  8
+#define DGEMM_DEFAULT_UNROLL_N  4
+
+#define CGEMM_DEFAULT_UNROLL_M  8
+#define CGEMM_DEFAULT_UNROLL_N  4
+
+#define ZGEMM_DEFAULT_UNROLL_M  4
+#define ZGEMM_DEFAULT_UNROLL_N  4
+
+#define SGEMM_DEFAULT_P 256
+#define DGEMM_DEFAULT_P 160
+#define CGEMM_DEFAULT_P 128
+#define ZGEMM_DEFAULT_P 128
+
+#define SGEMM_DEFAULT_Q 256
+#define DGEMM_DEFAULT_Q 128
+#define CGEMM_DEFAULT_Q 224
+#define ZGEMM_DEFAULT_Q 112
+
+#define SGEMM_DEFAULT_R 4096
+#define DGEMM_DEFAULT_R 4096
+#define CGEMM_DEFAULT_R 4096
+#define ZGEMM_DEFAULT_R 2048
+
 #elif defined(THUNDERX)
 
 #define SGEMM_DEFAULT_UNROLL_M  4
@@ -2977,7 +3028,7 @@ is a big desktop or server with abundant cache rather than a phone or embedded d
 #define GEMM_DEFAULT_OFFSET_B 0
 #define GEMM_DEFAULT_ALIGN 0x03fffUL
 
-#define SGEMM_DEFAULT_UNROLL_M  8
+#define SGEMM_DEFAULT_UNROLL_M  16
 #define SGEMM_DEFAULT_UNROLL_N  4
 
 #define DGEMM_DEFAULT_UNROLL_M  8
diff --git a/test/Makefile b/test/Makefile
index 074411b05..45f9821ec 100644
--- a/test/Makefile
+++ b/test/Makefile
@@ -1,7 +1,12 @@
 TOPDIR	= ..
 include ../Makefile.system
 
+
+ifeq ($(NOFORTRAN),1)
+all ::
+else
 all :: level1 level2 level3
+endif
 
 level1 : sblat1 dblat1 cblat1 zblat1
 ifndef CROSS
@@ -59,9 +64,17 @@ endif
 endif
 endif
 
+ifeq ($(BUILD_HALF),1)
+level3 : test_shgemm sblat3 dblat3 cblat3 zblat3
+else
 level3 : sblat3 dblat3 cblat3 zblat3
+endif
 ifndef CROSS
 	rm -f ?BLAT3.SUMM
+ifeq ($(BUILD_HALF),1)
+	OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 ./test_shgemm > SHBLAT3.SUMM
+	@$(GREP) -q FATAL SHBLAT3.SUMM && cat SHBLAT3.SUMM || exit 0
+endif
 	OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 ./sblat3 < ./sblat3.dat
 	@$(GREP) -q FATAL SBLAT3.SUMM && cat SBLAT3.SUMM || exit 0
 	OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 ./dblat3 < ./dblat3.dat
@@ -73,6 +86,10 @@ ifndef CROSS
 ifdef SMP
 	rm -f ?BLAT3.SUMM
 ifeq ($(USE_OPENMP), 1)
+ifeq ($(BUILD_HALF),1)
+	OMP_NUM_THREADS=2 ./test_shgemm > SHBLAT3.SUMM
+	@$(GREP) -q FATAL SHBLAT3.SUMM && cat SHBLAT3.SUMM || exit 0
+endif
 	OMP_NUM_THREADS=2 ./sblat3 < ./sblat3.dat
 	@$(GREP) -q FATAL SBLAT3.SUMM && cat SBLAT3.SUMM || exit 0
 	OMP_NUM_THREADS=2 ./dblat3 < ./dblat3.dat
@@ -82,6 +99,10 @@ ifeq ($(USE_OPENMP), 1)
 	OMP_NUM_THREADS=2 ./zblat3 < ./zblat3.dat
 	@$(GREP) -q FATAL ZBLAT3.SUMM && cat ZBLAT3.SUMM || exit 0
 else
+ifeq ($(BUILD_HALF),1)
+	OPENBLAS_NUM_THREADS=2 ./test_shgemm > SHBLAT3.SUMM
+	@$(GREP) -q FATAL SHBLAT3.SUMM && cat SHBLAT3.SUMM || exit 0
+endif
 	OPENBLAS_NUM_THREADS=2 ./sblat3 < ./sblat3.dat
 	@$(GREP) -q FATAL SBLAT3.SUMM && cat SBLAT3.SUMM || exit 0
 	OPENBLAS_NUM_THREADS=2 ./dblat3 < ./dblat3.dat
@@ -160,6 +181,11 @@ zblat2 : zblat2.$(SUFFIX) ../$(LIBNAME)
 sblat3 : sblat3.$(SUFFIX) ../$(LIBNAME)
 	$(FC) $(FLDFLAGS) -o sblat3 sblat3.$(SUFFIX) ../$(LIBNAME) $(EXTRALIB) $(CEXTRALIB)
 
+ifeq ($(BUILD_HALF),1)
+test_shgemm : compare_sgemm_shgemm.c ../$(LIBNAME)
+	$(FC) $(FLDFLAGS) -o test_shgemm compare_sgemm_shgemm.c ../$(LIBNAME) $(EXTRALIB) $(CEXTRALIB)
+endif
+
 dblat3 : dblat3.$(SUFFIX) ../$(LIBNAME)
 	$(FC) $(FLDFLAGS) -o dblat3 dblat3.$(SUFFIX) ../$(LIBNAME) $(EXTRALIB) $(CEXTRALIB)
 
@@ -182,7 +208,7 @@ clean:
 	@rm -f *.$(SUFFIX) *.$(PSUFFIX) gmon.$(SUFFIX)ut *.SUMM *.cxml *.exe *.pdb *.dwf \
 	sblat1 dblat1 cblat1 zblat1 \
 	sblat2 dblat2 cblat2 zblat2 \
-	sblat3 dblat3 cblat3 zblat3 \
+	test_shgemm sblat3 dblat3 cblat3 zblat3 \
 	sblat1p dblat1p cblat1p zblat1p \
 	sblat2p dblat2p cblat2p zblat2p \
 	sblat3p dblat3p cblat3p zblat3p \
diff --git a/test/cblat1.f b/test/cblat1.f
index d6b53d105..ecf2a44cb 100644
--- a/test/cblat1.f
+++ b/test/cblat1.f
@@ -1,7 +1,49 @@
+*> \brief \b CBLAT1
+*
+*  =========== DOCUMENTATION ===========
+*
+* Online html documentation available at
+*            http://www.netlib.org/lapack/explore-html/
+*
+*  Definition:
+*  ===========
+*
+*       PROGRAM CBLAT1
+*
+*
+*> \par Purpose:
+*  =============
+*>
+*> \verbatim
+*>
+*>    Test program for the COMPLEX Level 1 BLAS.
+*>    Based upon the original BLAS test routine together with:
+*>
+*>    F06GAF Example Program Text
+*> \endverbatim
+*
+*  Authors:
+*  ========
+*
+*> \author Univ. of Tennessee
+*> \author Univ. of California Berkeley
+*> \author Univ. of Colorado Denver
+*> \author NAG Ltd.
+*
+*> \date April 2012
+*
+*> \ingroup complex_blas_testing
+*
+*  =====================================================================
       PROGRAM CBLAT1
-*     Test program for the COMPLEX    Level 1 BLAS.
-*     Based upon the original BLAS test routine together with:
-*     F06GAF Example Program Text
+*
+*  -- Reference BLAS test routine (version 3.7.0) --
+*  -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
+*  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
+*     April 2012
+*
+*  =====================================================================
+*
 *     .. Parameters ..
       INTEGER          NOUT
       PARAMETER        (NOUT=6)
@@ -114,8 +156,8 @@
      +                  (5.0E0,6.0E0), (5.0E0,6.0E0), (0.1E0,0.1E0),
      +                  (-0.6E0,0.1E0), (0.1E0,-0.3E0), (7.0E0,8.0E0),
      +                  (7.0E0,8.0E0), (7.0E0,8.0E0), (7.0E0,8.0E0),
-     +                  (7.0E0,8.0E0), (0.3E0,0.1E0), (0.1E0,0.4E0),
-     +                  (0.4E0,0.1E0), (0.1E0,0.2E0), (2.0E0,3.0E0),
+     +                  (7.0E0,8.0E0), (0.3E0,0.1E0), (0.5E0,0.0E0),
+     +                  (0.0E0,0.5E0), (0.0E0,0.2E0), (2.0E0,3.0E0),
      +                  (2.0E0,3.0E0), (2.0E0,3.0E0), (2.0E0,3.0E0)/
       DATA              ((CV(I,J,2),I=1,8),J=1,5)/(0.1E0,0.1E0),
      +                  (4.0E0,5.0E0), (4.0E0,5.0E0), (4.0E0,5.0E0),
@@ -129,10 +171,10 @@
      +                  (3.0E0,6.0E0), (-0.6E0,0.1E0), (4.0E0,7.0E0),
      +                  (0.1E0,-0.3E0), (7.0E0,2.0E0), (7.0E0,2.0E0),
      +                  (7.0E0,2.0E0), (0.3E0,0.1E0), (5.0E0,8.0E0),
-     +                  (0.1E0,0.4E0), (6.0E0,9.0E0), (0.4E0,0.1E0),
-     +                  (8.0E0,3.0E0), (0.1E0,0.2E0), (9.0E0,4.0E0)/
-      DATA              STRUE2/0.0E0, 0.5E0, 0.6E0, 0.7E0, 0.7E0/
-      DATA              STRUE4/0.0E0, 0.7E0, 1.0E0, 1.3E0, 1.7E0/
+     +                  (0.5E0,0.0E0), (6.0E0,9.0E0), (0.0E0,0.5E0),
+     +                  (8.0E0,3.0E0), (0.0E0,0.2E0), (9.0E0,4.0E0)/
+      DATA              STRUE2/0.0E0, 0.5E0, 0.6E0, 0.7E0, 0.8E0/
+      DATA              STRUE4/0.0E0, 0.7E0, 1.0E0, 1.3E0, 1.6E0/
       DATA              ((CTRUE5(I,J,1),I=1,8),J=1,5)/(0.1E0,0.1E0),
      +                  (1.0E0,2.0E0), (1.0E0,2.0E0), (1.0E0,2.0E0),
      +                  (1.0E0,2.0E0), (1.0E0,2.0E0), (1.0E0,2.0E0),
@@ -145,8 +187,8 @@
      +                  (0.11E0,-0.03E0), (-0.17E0,0.46E0),
      +                  (-0.17E0,-0.19E0), (7.0E0,8.0E0), (7.0E0,8.0E0),
      +                  (7.0E0,8.0E0), (7.0E0,8.0E0), (7.0E0,8.0E0),
-     +                  (0.19E0,-0.17E0), (0.32E0,0.09E0),
-     +                  (0.23E0,-0.24E0), (0.18E0,0.01E0),
+     +                  (0.19E0,-0.17E0), (0.20E0,-0.35E0),
+     +                  (0.35E0,0.20E0), (0.14E0,0.08E0),
      +                  (2.0E0,3.0E0), (2.0E0,3.0E0), (2.0E0,3.0E0),
      +                  (2.0E0,3.0E0)/
       DATA              ((CTRUE5(I,J,2),I=1,8),J=1,5)/(0.1E0,0.1E0),
@@ -162,9 +204,9 @@
      +                  (-0.17E0,0.46E0), (4.0E0,7.0E0),
      +                  (-0.17E0,-0.19E0), (7.0E0,2.0E0), (7.0E0,2.0E0),
      +                  (7.0E0,2.0E0), (0.19E0,-0.17E0), (5.0E0,8.0E0),
-     +                  (0.32E0,0.09E0), (6.0E0,9.0E0),
-     +                  (0.23E0,-0.24E0), (8.0E0,3.0E0),
-     +                  (0.18E0,0.01E0), (9.0E0,4.0E0)/
+     +                  (0.20E0,-0.35E0), (6.0E0,9.0E0),
+     +                  (0.35E0,0.20E0), (8.0E0,3.0E0),
+     +                  (0.14E0,0.08E0), (9.0E0,4.0E0)/
       DATA              ((CTRUE6(I,J,1),I=1,8),J=1,5)/(0.1E0,0.1E0),
      +                  (1.0E0,2.0E0), (1.0E0,2.0E0), (1.0E0,2.0E0),
      +                  (1.0E0,2.0E0), (1.0E0,2.0E0), (1.0E0,2.0E0),
@@ -177,8 +219,8 @@
      +                  (0.03E0,0.03E0), (-0.18E0,0.03E0),
      +                  (0.03E0,-0.09E0), (7.0E0,8.0E0), (7.0E0,8.0E0),
      +                  (7.0E0,8.0E0), (7.0E0,8.0E0), (7.0E0,8.0E0),
-     +                  (0.09E0,0.03E0), (0.03E0,0.12E0),
-     +                  (0.12E0,0.03E0), (0.03E0,0.06E0), (2.0E0,3.0E0),
+     +                  (0.09E0,0.03E0), (0.15E0,0.00E0),
+     +                  (0.00E0,0.15E0), (0.00E0,0.06E0), (2.0E0,3.0E0),
      +                  (2.0E0,3.0E0), (2.0E0,3.0E0), (2.0E0,3.0E0)/
       DATA              ((CTRUE6(I,J,2),I=1,8),J=1,5)/(0.1E0,0.1E0),
      +                  (4.0E0,5.0E0), (4.0E0,5.0E0), (4.0E0,5.0E0),
@@ -193,8 +235,8 @@
      +                  (-0.18E0,0.03E0), (4.0E0,7.0E0),
      +                  (0.03E0,-0.09E0), (7.0E0,2.0E0), (7.0E0,2.0E0),
      +                  (7.0E0,2.0E0), (0.09E0,0.03E0), (5.0E0,8.0E0),
-     +                  (0.03E0,0.12E0), (6.0E0,9.0E0), (0.12E0,0.03E0),
-     +                  (8.0E0,3.0E0), (0.03E0,0.06E0), (9.0E0,4.0E0)/
+     +                  (0.15E0,0.00E0), (6.0E0,9.0E0), (0.00E0,0.15E0),
+     +                  (8.0E0,3.0E0), (0.00E0,0.06E0), (9.0E0,4.0E0)/
       DATA              ITRUE3/0, 1, 2, 2, 2/
 *     .. Executable Statements ..
       DO 60 INCX = 1, 2
@@ -529,7 +571,8 @@
 *
 *     .. Parameters ..
       INTEGER          NOUT
-      PARAMETER        (NOUT=6)
+      REAL             ZERO
+      PARAMETER        (NOUT=6, ZERO=0.0E0)
 *     .. Scalar Arguments ..
       REAL             SFAC
       INTEGER          LEN
@@ -552,7 +595,7 @@
 *
       DO 40 I = 1, LEN
          SD = SCOMP(I) - STRUE(I)
-         IF (SDIFF(ABS(SSIZE(I))+ABS(SFAC*SD),ABS(SSIZE(I))).EQ.0.0E0)
+         IF (ABS(SFAC*SD) .LE. ABS(SSIZE(I))*EPSILON(ZERO))
      +       GO TO 40
 *
 *                             HERE    SCOMP(I) IS NOT CLOSE TO STRUE(I).
diff --git a/test/cblat2.f b/test/cblat2.f
index 20f188100..8c7bac48e 100644
--- a/test/cblat2.f
+++ b/test/cblat2.f
@@ -1,68 +1,114 @@
+*> \brief \b CBLAT2
+*
+*  =========== DOCUMENTATION ===========
+*
+* Online html documentation available at
+*            http://www.netlib.org/lapack/explore-html/
+*
+*  Definition:
+*  ===========
+*
+*       PROGRAM CBLAT2
+*
+*
+*> \par Purpose:
+*  =============
+*>
+*> \verbatim
+*>
+*> Test program for the COMPLEX          Level 2 Blas.
+*>
+*> The program must be driven by a short data file. The first 18 records
+*> of the file are read using list-directed input, the last 17 records
+*> are read using the format ( A6, L2 ). An annotated example of a data
+*> file can be obtained by deleting the first 3 characters from the
+*> following 35 lines:
+*> 'cblat2.out'      NAME OF SUMMARY OUTPUT FILE
+*> 6                 UNIT NUMBER OF SUMMARY FILE
+*> 'CBLA2T.SNAP'     NAME OF SNAPSHOT OUTPUT FILE
+*> -1                UNIT NUMBER OF SNAPSHOT FILE (NOT USED IF .LT. 0)
+*> F        LOGICAL FLAG, T TO REWIND SNAPSHOT FILE AFTER EACH RECORD.
+*> F        LOGICAL FLAG, T TO STOP ON FAILURES.
+*> T        LOGICAL FLAG, T TO TEST ERROR EXITS.
+*> 16.0     THRESHOLD VALUE OF TEST RATIO
+*> 6                 NUMBER OF VALUES OF N
+*> 0 1 2 3 5 9       VALUES OF N
+*> 4                 NUMBER OF VALUES OF K
+*> 0 1 2 4           VALUES OF K
+*> 4                 NUMBER OF VALUES OF INCX AND INCY
+*> 1 2 -1 -2         VALUES OF INCX AND INCY
+*> 3                 NUMBER OF VALUES OF ALPHA
+*> (0.0,0.0) (1.0,0.0) (0.7,-0.9)       VALUES OF ALPHA
+*> 3                 NUMBER OF VALUES OF BETA
+*> (0.0,0.0) (1.0,0.0) (1.3,-1.1)       VALUES OF BETA
+*> CGEMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CGBMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CHEMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CHBMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CHPMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CTRMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CTBMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CTPMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CTRSV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CTBSV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CTPSV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CGERC  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CGERU  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CHER   T PUT F FOR NO TEST. SAME COLUMNS.
+*> CHPR   T PUT F FOR NO TEST. SAME COLUMNS.
+*> CHER2  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CHPR2  T PUT F FOR NO TEST. SAME COLUMNS.
+*>
+*> Further Details
+*> ===============
+*>
+*>    See:
+*>
+*>       Dongarra J. J., Du Croz J. J., Hammarling S.  and Hanson R. J..
+*>       An  extended  set of Fortran  Basic Linear Algebra Subprograms.
+*>
+*>       Technical  Memoranda  Nos. 41 (revision 3) and 81,  Mathematics
+*>       and  Computer Science  Division,  Argonne  National Laboratory,
+*>       9700 South Cass Avenue, Argonne, Illinois 60439, US.
+*>
+*>       Or
+*>
+*>       NAG  Technical Reports TR3/87 and TR4/87,  Numerical Algorithms
+*>       Group  Ltd.,  NAG  Central  Office,  256  Banbury  Road, Oxford
+*>       OX2 7DE, UK,  and  Numerical Algorithms Group Inc.,  1101  31st
+*>       Street,  Suite 100,  Downers Grove,  Illinois 60515-1263,  USA.
+*>
+*>
+*> -- Written on 10-August-1987.
+*>    Richard Hanson, Sandia National Labs.
+*>    Jeremy Du Croz, NAG Central Office.
+*>
+*>    10-9-00:  Change STATUS='NEW' to 'UNKNOWN' so that the testers
+*>              can be run multiple times without deleting generated
+*>              output files (susan)
+*> \endverbatim
+*
+*  Authors:
+*  ========
+*
+*> \author Univ. of Tennessee
+*> \author Univ. of California Berkeley
+*> \author Univ. of Colorado Denver
+*> \author NAG Ltd.
+*
+*> \date April 2012
+*
+*> \ingroup complex_blas_testing
+*
+*  =====================================================================
       PROGRAM CBLAT2
 *
-*  Test program for the COMPLEX          Level 2 Blas.
+*  -- Reference BLAS test routine (version 3.7.0) --
+*  -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
+*  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
+*     April 2012
 *
-*  The program must be driven by a short data file. The first 18 records
-*  of the file are read using list-directed input, the last 17 records
-*  are read using the format ( A6, L2 ). An annotated example of a data
-*  file can be obtained by deleting the first 3 characters from the
-*  following 35 lines:
-*  'CBLAT2.SUMM'     NAME OF SUMMARY OUTPUT FILE
-*  6                 UNIT NUMBER OF SUMMARY FILE
-*  'CBLA2T.SNAP'     NAME OF SNAPSHOT OUTPUT FILE
-*  -1                UNIT NUMBER OF SNAPSHOT FILE (NOT USED IF .LT. 0)
-*  F        LOGICAL FLAG, T TO REWIND SNAPSHOT FILE AFTER EACH RECORD.
-*  F        LOGICAL FLAG, T TO STOP ON FAILURES.
-*  T        LOGICAL FLAG, T TO TEST ERROR EXITS.
-*  16.0     THRESHOLD VALUE OF TEST RATIO
-*  6                 NUMBER OF VALUES OF N
-*  0 1 2 3 5 9       VALUES OF N
-*  4                 NUMBER OF VALUES OF K
-*  0 1 2 4           VALUES OF K
-*  4                 NUMBER OF VALUES OF INCX AND INCY
-*  1 2 -1 -2         VALUES OF INCX AND INCY
-*  3                 NUMBER OF VALUES OF ALPHA
-*  (0.0,0.0) (1.0,0.0) (0.7,-0.9)       VALUES OF ALPHA
-*  3                 NUMBER OF VALUES OF BETA
-*  (0.0,0.0) (1.0,0.0) (1.3,-1.1)       VALUES OF BETA
-*  CGEMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CGBMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CHEMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CHBMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CHPMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CTRMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CTBMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CTPMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CTRSV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CTBSV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CTPSV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CGERC  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CGERU  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CHER   T PUT F FOR NO TEST. SAME COLUMNS.
-*  CHPR   T PUT F FOR NO TEST. SAME COLUMNS.
-*  CHER2  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CHPR2  T PUT F FOR NO TEST. SAME COLUMNS.
-*
-*     See:
-*
-*        Dongarra J. J., Du Croz J. J., Hammarling S.  and Hanson R. J..
-*        An  extended  set of Fortran  Basic Linear Algebra Subprograms.
-*
-*        Technical  Memoranda  Nos. 41 (revision 3) and 81,  Mathematics
-*        and  Computer Science  Division,  Argonne  National Laboratory,
-*        9700 South Cass Avenue, Argonne, Illinois 60439, US.
-*
-*        Or
-*
-*        NAG  Technical Reports TR3/87 and TR4/87,  Numerical Algorithms
-*        Group  Ltd.,  NAG  Central  Office,  256  Banbury  Road, Oxford
-*        OX2 7DE, UK,  and  Numerical Algorithms Group Inc.,  1101  31st
-*        Street,  Suite 100,  Downers Grove,  Illinois 60515-1263,  USA.
-*
-*
-*  -- Written on 10-August-1987.
-*     Richard Hanson, Sandia National Labs.
-*     Jeremy Du Croz, NAG Central Office.
+*  =====================================================================
 *
 *     .. Parameters ..
       INTEGER            NIN
@@ -71,8 +117,8 @@
       PARAMETER          ( NSUBS = 17 )
       COMPLEX            ZERO, ONE
       PARAMETER          ( ZERO = ( 0.0, 0.0 ), ONE = ( 1.0, 0.0 ) )
-      REAL               RZERO, RHALF, RONE
-      PARAMETER          ( RZERO = 0.0, RHALF = 0.5, RONE = 1.0 )
+      REAL               RZERO
+      PARAMETER          ( RZERO = 0.0 )
       INTEGER            NMAX, INCMAX
       PARAMETER          ( NMAX = 65, INCMAX = 2 )
       INTEGER            NINMAX, NIDMAX, NKBMAX, NALMAX, NBEMAX
@@ -126,7 +172,7 @@
 *
       READ( NIN, FMT = * )SUMMRY
       READ( NIN, FMT = * )NOUT
-      OPEN( NOUT, FILE = SUMMRY, STATUS = 'NEW' )
+      OPEN( NOUT, FILE = SUMMRY, STATUS = 'UNKNOWN' )
       NOUTC = NOUT
 *
 *     Read name and unit number for snapshot output file and open file.
@@ -135,7 +181,7 @@
       READ( NIN, FMT = * )NTRA
       TRACE = NTRA.GE.0
       IF( TRACE )THEN
-         OPEN( NTRA, FILE = SNAPS, STATUS = 'NEW' )
+         OPEN( NTRA, FILE = SNAPS, STATUS = 'UNKNOWN' )
       END IF
 *     Read the flag that directs rewinding of the snapshot file.
       READ( NIN, FMT = * )REWI
@@ -240,14 +286,7 @@
 *
 *     Compute EPS (the machine precision).
 *
-      EPS = RONE
-   90 CONTINUE
-      IF( SDIFF( RONE + EPS, RONE ).EQ.RZERO )
-     $   GO TO 100
-      EPS = RHALF*EPS
-      GO TO 90
-  100 CONTINUE
-      EPS = EPS + EPS
+      EPS = EPSILON(RZERO)
       WRITE( NOUT, FMT = 9998 )EPS
 *
 *     Check the reliability of CMVCH using exact data.
@@ -3079,7 +3118,6 @@
    50    CONTINUE
       END IF
 *
-   60 CONTINUE
       LCERES = .TRUE.
       GO TO 80
    70 CONTINUE
diff --git a/test/cblat3.f b/test/cblat3.f
index 5df1ddd64..a65e1364c 100644
--- a/test/cblat3.f
+++ b/test/cblat3.f
@@ -1,50 +1,96 @@
+*> \brief \b CBLAT3
+*
+*  =========== DOCUMENTATION ===========
+*
+* Online html documentation available at
+*            http://www.netlib.org/lapack/explore-html/
+*
+*  Definition:
+*  ===========
+*
+*       PROGRAM CBLAT3
+*
+*
+*> \par Purpose:
+*  =============
+*>
+*> \verbatim
+*>
+*> Test program for the COMPLEX          Level 3 Blas.
+*>
+*> The program must be driven by a short data file. The first 14 records
+*> of the file are read using list-directed input, the last 9 records
+*> are read using the format ( A6, L2 ). An annotated example of a data
+*> file can be obtained by deleting the first 3 characters from the
+*> following 23 lines:
+*> 'cblat3.out'      NAME OF SUMMARY OUTPUT FILE
+*> 6                 UNIT NUMBER OF SUMMARY FILE
+*> 'CBLAT3.SNAP'     NAME OF SNAPSHOT OUTPUT FILE
+*> -1                UNIT NUMBER OF SNAPSHOT FILE (NOT USED IF .LT. 0)
+*> F        LOGICAL FLAG, T TO REWIND SNAPSHOT FILE AFTER EACH RECORD.
+*> F        LOGICAL FLAG, T TO STOP ON FAILURES.
+*> T        LOGICAL FLAG, T TO TEST ERROR EXITS.
+*> 16.0     THRESHOLD VALUE OF TEST RATIO
+*> 6                 NUMBER OF VALUES OF N
+*> 0 1 2 3 5 9       VALUES OF N
+*> 3                 NUMBER OF VALUES OF ALPHA
+*> (0.0,0.0) (1.0,0.0) (0.7,-0.9)       VALUES OF ALPHA
+*> 3                 NUMBER OF VALUES OF BETA
+*> (0.0,0.0) (1.0,0.0) (1.3,-1.1)       VALUES OF BETA
+*> CGEMM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CHEMM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CSYMM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CTRMM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CTRSM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CHERK  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CSYRK  T PUT F FOR NO TEST. SAME COLUMNS.
+*> CHER2K T PUT F FOR NO TEST. SAME COLUMNS.
+*> CSYR2K T PUT F FOR NO TEST. SAME COLUMNS.
+*>
+*> Further Details
+*> ===============
+*>
+*> See:
+*>
+*>    Dongarra J. J., Du Croz J. J., Duff I. S. and Hammarling S.
+*>    A Set of Level 3 Basic Linear Algebra Subprograms.
+*>
+*>    Technical Memorandum No.88 (Revision 1), Mathematics and
+*>    Computer Science Division, Argonne National Laboratory, 9700
+*>    South Cass Avenue, Argonne, Illinois 60439, US.
+*>
+*> -- Written on 8-February-1989.
+*>    Jack Dongarra, Argonne National Laboratory.
+*>    Iain Duff, AERE Harwell.
+*>    Jeremy Du Croz, Numerical Algorithms Group Ltd.
+*>    Sven Hammarling, Numerical Algorithms Group Ltd.
+*>
+*>    10-9-00:  Change STATUS='NEW' to 'UNKNOWN' so that the testers
+*>              can be run multiple times without deleting generated
+*>              output files (susan)
+*> \endverbatim
+*
+*  Authors:
+*  ========
+*
+*> \author Univ. of Tennessee
+*> \author Univ. of California Berkeley
+*> \author Univ. of Colorado Denver
+*> \author NAG Ltd.
+*
+*> \date April 2012
+*
+*> \ingroup complex_blas_testing
+*
+*  =====================================================================
       PROGRAM CBLAT3
 *
-*  Test program for the COMPLEX          Level 3 Blas.
+*  -- Reference BLAS test routine (version 3.7.0) --
+*  -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
+*  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
+*     April 2012
 *
-*  The program must be driven by a short data file. The first 14 records
-*  of the file are read using list-directed input, the last 9 records
-*  are read using the format ( A6, L2 ). An annotated example of a data
-*  file can be obtained by deleting the first 3 characters from the
-*  following 23 lines:
-*  'CBLAT3.SUMM'     NAME OF SUMMARY OUTPUT FILE
-*  6                 UNIT NUMBER OF SUMMARY FILE
-*  'CBLAT3.SNAP'     NAME OF SNAPSHOT OUTPUT FILE
-*  -1                UNIT NUMBER OF SNAPSHOT FILE (NOT USED IF .LT. 0)
-*  F        LOGICAL FLAG, T TO REWIND SNAPSHOT FILE AFTER EACH RECORD.
-*  F        LOGICAL FLAG, T TO STOP ON FAILURES.
-*  T        LOGICAL FLAG, T TO TEST ERROR EXITS.
-*  16.0     THRESHOLD VALUE OF TEST RATIO
-*  6                 NUMBER OF VALUES OF N
-*  0 1 2 3 5 9       VALUES OF N
-*  3                 NUMBER OF VALUES OF ALPHA
-*  (0.0,0.0) (1.0,0.0) (0.7,-0.9)       VALUES OF ALPHA
-*  3                 NUMBER OF VALUES OF BETA
-*  (0.0,0.0) (1.0,0.0) (1.3,-1.1)       VALUES OF BETA
-*  CGEMM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CHEMM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CSYMM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CTRMM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CTRSM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CHERK  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CSYRK  T PUT F FOR NO TEST. SAME COLUMNS.
-*  CHER2K T PUT F FOR NO TEST. SAME COLUMNS.
-*  CSYR2K T PUT F FOR NO TEST. SAME COLUMNS.
-*
-*  See:
-*
-*     Dongarra J. J., Du Croz J. J., Duff I. S. and Hammarling S.
-*     A Set of Level 3 Basic Linear Algebra Subprograms.
-*
-*     Technical Memorandum No.88 (Revision 1), Mathematics and
-*     Computer Science Division, Argonne National Laboratory, 9700
-*     South Cass Avenue, Argonne, Illinois 60439, US.
-*
-*  -- Written on 8-February-1989.
-*     Jack Dongarra, Argonne National Laboratory.
-*     Iain Duff, AERE Harwell.
-*     Jeremy Du Croz, Numerical Algorithms Group Ltd.
-*     Sven Hammarling, Numerical Algorithms Group Ltd.
+*  =====================================================================
 *
 *     .. Parameters ..
       INTEGER            NIN
@@ -53,8 +99,8 @@
       PARAMETER          ( NSUBS = 9 )
       COMPLEX            ZERO, ONE
       PARAMETER          ( ZERO = ( 0.0, 0.0 ), ONE = ( 1.0, 0.0 ) )
-      REAL               RZERO, RHALF, RONE
-      PARAMETER          ( RZERO = 0.0, RHALF = 0.5, RONE = 1.0 )
+      REAL               RZERO
+      PARAMETER          ( RZERO = 0.0 )
       INTEGER            NMAX
       PARAMETER          ( NMAX = 65 )
       INTEGER            NIDMAX, NALMAX, NBEMAX
@@ -103,7 +149,7 @@
 *
       READ( NIN, FMT = * )SUMMRY
       READ( NIN, FMT = * )NOUT
-      OPEN( NOUT, FILE = SUMMRY, STATUS = 'NEW' )
+      OPEN( NOUT, FILE = SUMMRY )
       NOUTC = NOUT
 *
 *     Read name and unit number for snapshot output file and open file.
@@ -112,7 +158,7 @@
       READ( NIN, FMT = * )NTRA
       TRACE = NTRA.GE.0
       IF( TRACE )THEN
-         OPEN( NTRA, FILE = SNAPS, STATUS = 'NEW' )
+         OPEN( NTRA, FILE = SNAPS )
       END IF
 *     Read the flag that directs rewinding of the snapshot file.
       READ( NIN, FMT = * )REWI
@@ -189,14 +235,7 @@
 *
 *     Compute EPS (the machine precision).
 *
-      EPS = RONE
-   70 CONTINUE
-      IF( SDIFF( RONE + EPS, RONE ).EQ.RZERO )
-     $   GO TO 80
-      EPS = RHALF*EPS
-      GO TO 70
-   80 CONTINUE
-      EPS = EPS + EPS
+      EPS = EPSILON(RZERO)
       WRITE( NOUT, FMT = 9998 )EPS
 *
 *     Check the reliability of CMMCH using exact data.
@@ -1301,8 +1340,6 @@
       NC = 0
       RESET = .TRUE.
       ERRMAX = RZERO
-      RALS = RONE
-      RBETS = RONE
 *
       DO 100 IN = 1, NIDIM
          N = IDIM( IN )
@@ -1948,7 +1985,7 @@
 *
 *  Tests the error exits from the Level 3 Blas.
 *  Requires a special version of the error-handling routine XERBLA.
-*  ALPHA, RALPHA, BETA, RBETA, A, B and C should not need to be defined.
+*  A, B and C should not need to be defined.
 *
 *  Auxiliary routine for test program for Level 3 Blas.
 *
@@ -1958,12 +1995,19 @@
 *     Jeremy Du Croz, Numerical Algorithms Group Ltd.
 *     Sven Hammarling, Numerical Algorithms Group Ltd.
 *
+*  3-19-92:  Initialize ALPHA, BETA, RALPHA, and RBETA  (eca)
+*  3-19-92:  Fix argument 12 in calls to CSYMM and CHEMM
+*            with INFOT = 9  (eca)
+*
 *     .. Scalar Arguments ..
       INTEGER            ISNUM, NOUT
       CHARACTER*6        SRNAMT
 *     .. Scalars in Common ..
       INTEGER            INFOT, NOUTC
       LOGICAL            LERR, OK
+*     .. Parameters ..
+      REAL               ONE, TWO
+      PARAMETER          ( ONE = 1.0E0, TWO = 2.0E0 )
 *     .. Local Scalars ..
       COMPLEX            ALPHA, BETA
       REAL               RALPHA, RBETA
@@ -1981,6 +2025,14 @@
 *     LERR is set to .TRUE. by the special version of XERBLA each time
 *     it is called, and is then tested and re-set by CHKXER.
       LERR = .FALSE.
+*
+*     Initialize ALPHA, BETA, RALPHA, and RBETA.
+*
+      ALPHA = CMPLX( ONE, -ONE )
+      BETA = CMPLX( TWO, -TWO )
+      RALPHA = ONE
+      RBETA = TWO
+*
       GO TO ( 10, 20, 30, 40, 50, 60, 70, 80,
      $        90 )ISNUM
    10 INFOT = 1
@@ -2207,16 +2259,16 @@
       CALL CHEMM( 'R', 'L', 0, 2, ALPHA, A, 1, B, 1, BETA, C, 1 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL CHEMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 1 )
+      CALL CHEMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL CHEMM( 'R', 'U', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 1 )
+      CALL CHEMM( 'R', 'U', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL CHEMM( 'L', 'L', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 1 )
+      CALL CHEMM( 'L', 'L', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL CHEMM( 'R', 'L', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 1 )
+      CALL CHEMM( 'R', 'L', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 12
       CALL CHEMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 2, BETA, C, 1 )
@@ -2274,16 +2326,16 @@
       CALL CSYMM( 'R', 'L', 0, 2, ALPHA, A, 1, B, 1, BETA, C, 1 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL CSYMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 1 )
+      CALL CSYMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL CSYMM( 'R', 'U', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 1 )
+      CALL CSYMM( 'R', 'U', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL CSYMM( 'L', 'L', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 1 )
+      CALL CSYMM( 'L', 'L', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL CSYMM( 'R', 'L', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 1 )
+      CALL CSYMM( 'R', 'L', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 12
       CALL CSYMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 2, BETA, C, 1 )
@@ -3270,7 +3322,6 @@
    50    CONTINUE
       END IF
 *
-   60 CONTINUE
       LCERES = .TRUE.
       GO TO 80
    70 CONTINUE
diff --git a/test/compare_sgemm_shgemm.c b/test/compare_sgemm_shgemm.c
new file mode 100644
index 000000000..57aee7b8f
--- /dev/null
+++ b/test/compare_sgemm_shgemm.c
@@ -0,0 +1,131 @@
+/***************************************************************************
+Copyright (c) 2020, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+#include <stdio.h>
+#include <stdint.h>
+#include "../common.h"
+#define SGEMM   BLASFUNC(sgemm)
+#define SHGEMM   BLASFUNC(shgemm)
+typedef union
+{
+  unsigned short v;
+  struct
+  {
+#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+    unsigned short s:1;
+    unsigned short e:8;
+    unsigned short m:7;
+#else
+    unsigned short m:7;
+    unsigned short e:8;
+    unsigned short s:1;
+#endif
+  } bits;
+} bfloat16_bits;
+
+typedef union
+{
+  float v;
+  struct
+  {
+    uint32_t m:23;
+    uint32_t e:8;
+    uint32_t s:1;
+  } bits;
+} float32_bits;
+
+float
+float16to32 (bfloat16_bits f16)
+{
+  float32_bits f32;
+  f32.bits.s = f16.bits.s;
+  f32.bits.e = f16.bits.e;
+  f32.bits.m = (uint32_t) f16.bits.m << 16;
+  return f32.v;
+}
+
+int
+main (int argc, char *argv[])
+{
+  int m, n, k;
+  int i, j, l;
+  int x;
+  int ret = 0;
+  int loop = 100;
+  char transA = 'N', transB = 'N';
+  float alpha = 1.0, beta = 0.0;
+
+  for (x = 0; x <= loop; x++)
+    {
+      m = k = n = x;
+      float A[m * k];
+      float B[k * n];
+      float C[m * n];
+      bfloat16_bits AA[m * k], BB[k * n];
+      float DD[m * n], CC[m * n];
+
+      for (j = 0; j < m; j++)
+	{
+	  for (i = 0; i < m; i++)
+	    {
+	      A[j * k + i] = ((FLOAT) rand () / (FLOAT) RAND_MAX) + 0.5;
+	      B[j * k + i] = ((FLOAT) rand () / (FLOAT) RAND_MAX) + 0.5;
+	      C[j * k + i] = 0;
+	      AA[j * k + i].v = *(uint32_t *) & A[j * k + i] >> 16;
+	      BB[j * k + i].v = *(uint32_t *) & B[j * k + i] >> 16;
+	      CC[j * k + i] = 0;
+	      DD[j * k + i] = 0;
+	    }
+	}
+      SGEMM (&transA, &transB, &m, &n, &k, &alpha, A,
+	     &m, B, &k, &beta, C, &m);
+      SHGEMM (&transA, &transB, &m, &n, &k, &alpha, AA,
+	      &m, BB, &k, &beta, CC, &m);
+      for (i = 0; i < n; i++)
+	for (j = 0; j < m; j++)
+	  for (l = 0; l < k; l++)
+	    if (fabs (CC[i * m + j] - C[i * m + j]) > 1.0)
+	      ret++;
+      if (transA == 'N' && transB == 'N')
+	{
+	  for (i = 0; i < n; i++)
+	    for (j = 0; j < m; j++)
+	      for (l = 0; l < k; l++)
+		{
+		  DD[i * m + j] +=
+		    float16to32 (AA[l * m + j]) * float16to32 (BB[l + k * i]);
+		}
+	  for (i = 0; i < n; i++)
+	    for (j = 0; j < m; j++)
+	      for (l = 0; l < k; l++)
+		if (CC[i * m + j] != DD[i * m + j])
+		  ret++;
+	}
+    }
+  if (ret != 0)
+    fprintf (stderr, "FATAL ERROR SHGEMM - Return code: %d\n", ret);
+  return ret;
+}
diff --git a/test/dblat2.f b/test/dblat2.f
index 4002d4368..9bbbe9792 100644
--- a/test/dblat2.f
+++ b/test/dblat2.f
@@ -1,75 +1,121 @@
+*> \brief \b DBLAT2
+*
+*  =========== DOCUMENTATION ===========
+*
+* Online html documentation available at
+*            http://www.netlib.org/lapack/explore-html/
+*
+*  Definition:
+*  ===========
+*
+*       PROGRAM DBLAT2
+*
+*
+*> \par Purpose:
+*  =============
+*>
+*> \verbatim
+*>
+*> Test program for the DOUBLE PRECISION Level 2 Blas.
+*>
+*> The program must be driven by a short data file. The first 18 records
+*> of the file are read using list-directed input, the last 16 records
+*> are read using the format ( A6, L2 ). An annotated example of a data
+*> file can be obtained by deleting the first 3 characters from the
+*> following 34 lines:
+*> 'dblat2.out'      NAME OF SUMMARY OUTPUT FILE
+*> 6                 UNIT NUMBER OF SUMMARY FILE
+*> 'DBLAT2.SNAP'     NAME OF SNAPSHOT OUTPUT FILE
+*> -1                UNIT NUMBER OF SNAPSHOT FILE (NOT USED IF .LT. 0)
+*> F        LOGICAL FLAG, T TO REWIND SNAPSHOT FILE AFTER EACH RECORD.
+*> F        LOGICAL FLAG, T TO STOP ON FAILURES.
+*> T        LOGICAL FLAG, T TO TEST ERROR EXITS.
+*> 16.0     THRESHOLD VALUE OF TEST RATIO
+*> 6                 NUMBER OF VALUES OF N
+*> 0 1 2 3 5 9       VALUES OF N
+*> 4                 NUMBER OF VALUES OF K
+*> 0 1 2 4           VALUES OF K
+*> 4                 NUMBER OF VALUES OF INCX AND INCY
+*> 1 2 -1 -2         VALUES OF INCX AND INCY
+*> 3                 NUMBER OF VALUES OF ALPHA
+*> 0.0 1.0 0.7       VALUES OF ALPHA
+*> 3                 NUMBER OF VALUES OF BETA
+*> 0.0 1.0 0.9       VALUES OF BETAC
+*> DGEMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> DGBMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> DSYMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> DSBMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> DSPMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> DTRMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> DTBMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> DTPMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> DTRSV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> DTBSV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> DTPSV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> DGER   T PUT F FOR NO TEST. SAME COLUMNS.
+*> DSYR   T PUT F FOR NO TEST. SAME COLUMNS.
+*> DSPR   T PUT F FOR NO TEST. SAME COLUMNS.
+*> DSYR2  T PUT F FOR NO TEST. SAME COLUMNS.
+*> DSPR2  T PUT F FOR NO TEST. SAME COLUMNS.
+*>
+*> Further Details
+*> ===============
+*>
+*>    See:
+*>
+*>       Dongarra J. J., Du Croz J. J., Hammarling S.  and Hanson R. J..
+*>       An  extended  set of Fortran  Basic Linear Algebra Subprograms.
+*>
+*>       Technical  Memoranda  Nos. 41 (revision 3) and 81,  Mathematics
+*>       and  Computer Science  Division,  Argonne  National Laboratory,
+*>       9700 South Cass Avenue, Argonne, Illinois 60439, US.
+*>
+*>       Or
+*>
+*>       NAG  Technical Reports TR3/87 and TR4/87,  Numerical Algorithms
+*>       Group  Ltd.,  NAG  Central  Office,  256  Banbury  Road, Oxford
+*>       OX2 7DE, UK,  and  Numerical Algorithms Group Inc.,  1101  31st
+*>       Street,  Suite 100,  Downers Grove,  Illinois 60515-1263,  USA.
+*>
+*>
+*> -- Written on 10-August-1987.
+*>    Richard Hanson, Sandia National Labs.
+*>    Jeremy Du Croz, NAG Central Office.
+*>
+*>    10-9-00:  Change STATUS='NEW' to 'UNKNOWN' so that the testers
+*>              can be run multiple times without deleting generated
+*>              output files (susan)
+*> \endverbatim
+*
+*  Authors:
+*  ========
+*
+*> \author Univ. of Tennessee
+*> \author Univ. of California Berkeley
+*> \author Univ. of Colorado Denver
+*> \author NAG Ltd.
+*
+*> \date April 2012
+*
+*> \ingroup double_blas_testing
+*
+*  =====================================================================
       PROGRAM DBLAT2
 *
-*  Test program for the DOUBLE PRECISION Level 2 Blas.
+*  -- Reference BLAS test routine (version 3.7.0) --
+*  -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
+*  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
+*     April 2012
 *
-*  The program must be driven by a short data file. The first 18 records
-*  of the file are read using list-directed input, the last 16 records
-*  are read using the format ( A6, L2 ). An annotated example of a data
-*  file can be obtained by deleting the first 3 characters from the
-*  following 34 lines:
-*  'DBLAT2.SUMM'     NAME OF SUMMARY OUTPUT FILE
-*  6                 UNIT NUMBER OF SUMMARY FILE
-*  'DBLAT2.SNAP'     NAME OF SNAPSHOT OUTPUT FILE
-*  -1                UNIT NUMBER OF SNAPSHOT FILE (NOT USED IF .LT. 0)
-*  F        LOGICAL FLAG, T TO REWIND SNAPSHOT FILE AFTER EACH RECORD.
-*  F        LOGICAL FLAG, T TO STOP ON FAILURES.
-*  T        LOGICAL FLAG, T TO TEST ERROR EXITS.
-*  16.0     THRESHOLD VALUE OF TEST RATIO
-*  6                 NUMBER OF VALUES OF N
-*  0 1 2 3 5 9       VALUES OF N
-*  4                 NUMBER OF VALUES OF K
-*  0 1 2 4           VALUES OF K
-*  4                 NUMBER OF VALUES OF INCX AND INCY
-*  1 2 -1 -2         VALUES OF INCX AND INCY
-*  3                 NUMBER OF VALUES OF ALPHA
-*  0.0 1.0 0.7       VALUES OF ALPHA
-*  3                 NUMBER OF VALUES OF BETA
-*  0.0 1.0 0.9       VALUES OF BETA
-*  DGEMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  DGBMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  DSYMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  DSBMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  DSPMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  DTRMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  DTBMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  DTPMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  DTRSV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  DTBSV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  DTPSV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  DGER   T PUT F FOR NO TEST. SAME COLUMNS.
-*  DSYR   T PUT F FOR NO TEST. SAME COLUMNS.
-*  DSPR   T PUT F FOR NO TEST. SAME COLUMNS.
-*  DSYR2  T PUT F FOR NO TEST. SAME COLUMNS.
-*  DSPR2  T PUT F FOR NO TEST. SAME COLUMNS.
-*
-*     See:
-*
-*        Dongarra J. J., Du Croz J. J., Hammarling S.  and Hanson R. J..
-*        An  extended  set of Fortran  Basic Linear Algebra Subprograms.
-*
-*        Technical  Memoranda  Nos. 41 (revision 3) and 81,  Mathematics
-*        and  Computer Science  Division,  Argonne  National Laboratory,
-*        9700 South Cass Avenue, Argonne, Illinois 60439, US.
-*
-*        Or
-*
-*        NAG  Technical Reports TR3/87 and TR4/87,  Numerical Algorithms
-*        Group  Ltd.,  NAG  Central  Office,  256  Banbury  Road, Oxford
-*        OX2 7DE, UK,  and  Numerical Algorithms Group Inc.,  1101  31st
-*        Street,  Suite 100,  Downers Grove,  Illinois 60515-1263,  USA.
-*
-*
-*  -- Written on 10-August-1987.
-*     Richard Hanson, Sandia National Labs.
-*     Jeremy Du Croz, NAG Central Office.
+*  =====================================================================
 *
 *     .. Parameters ..
       INTEGER            NIN
       PARAMETER          ( NIN = 5 )
       INTEGER            NSUBS
       PARAMETER          ( NSUBS = 16 )
-      DOUBLE PRECISION   ZERO, HALF, ONE
-      PARAMETER          ( ZERO = 0.0D0, HALF = 0.5D0, ONE = 1.0D0 )
+      DOUBLE PRECISION   ZERO, ONE
+      PARAMETER          ( ZERO = 0.0D0, ONE = 1.0D0 )
       INTEGER            NMAX, INCMAX
       PARAMETER          ( NMAX = 65, INCMAX = 2 )
       INTEGER            NINMAX, NIDMAX, NKBMAX, NALMAX, NBEMAX
@@ -121,7 +167,7 @@
 *
       READ( NIN, FMT = * )SUMMRY
       READ( NIN, FMT = * )NOUT
-      OPEN( NOUT, FILE = SUMMRY, STATUS = 'NEW' )
+      OPEN( NOUT, FILE = SUMMRY, STATUS = 'UNKNOWN' )
       NOUTC = NOUT
 *
 *     Read name and unit number for snapshot output file and open file.
@@ -130,7 +176,7 @@
       READ( NIN, FMT = * )NTRA
       TRACE = NTRA.GE.0
       IF( TRACE )THEN
-         OPEN( NTRA, FILE = SNAPS, STATUS = 'NEW' )
+         OPEN( NTRA, FILE = SNAPS, STATUS = 'UNKNOWN' )
       END IF
 *     Read the flag that directs rewinding of the snapshot file.
       READ( NIN, FMT = * )REWI
@@ -235,14 +281,7 @@
 *
 *     Compute EPS (the machine precision).
 *
-      EPS = ONE
-   90 CONTINUE
-      IF( DDIFF( ONE + EPS, ONE ).EQ.ZERO )
-     $   GO TO 100
-      EPS = HALF*EPS
-      GO TO 90
-  100 CONTINUE
-      EPS = EPS + EPS
+      EPS = EPSILON(ZERO)
       WRITE( NOUT, FMT = 9998 )EPS
 *
 *     Check the reliability of DMVCH using exact data.
@@ -2982,7 +3021,6 @@
    50    CONTINUE
       END IF
 *
-   60 CONTINUE
       LDERES = .TRUE.
       GO TO 80
    70 CONTINUE
diff --git a/test/dblat3.f b/test/dblat3.f
index 082e03e5e..1ebec4ffa 100644
--- a/test/dblat3.f
+++ b/test/dblat3.f
@@ -1,55 +1,101 @@
+*> \brief \b DBLAT3
+*
+*  =========== DOCUMENTATION ===========
+*
+* Online html documentation available at
+*            http://www.netlib.org/lapack/explore-html/
+*
+*  Definition:
+*  ===========
+*
+*       PROGRAM DBLAT3
+*
+*
+*> \par Purpose:
+*  =============
+*>
+*> \verbatim
+*>
+*> Test program for the DOUBLE PRECISION Level 3 Blas.
+*>
+*> The program must be driven by a short data file. The first 14 records
+*> of the file are read using list-directed input, the last 6 records
+*> are read using the format ( A6, L2 ). An annotated example of a data
+*> file can be obtained by deleting the first 3 characters from the
+*> following 20 lines:
+*> 'dblat3.out'      NAME OF SUMMARY OUTPUT FILE
+*> 6                 UNIT NUMBER OF SUMMARY FILE
+*> 'DBLAT3.SNAP'     NAME OF SNAPSHOT OUTPUT FILE
+*> -1                UNIT NUMBER OF SNAPSHOT FILE (NOT USED IF .LT. 0)
+*> F        LOGICAL FLAG, T TO REWIND SNAPSHOT FILE AFTER EACH RECORD.
+*> F        LOGICAL FLAG, T TO STOP ON FAILURES.
+*> T        LOGICAL FLAG, T TO TEST ERROR EXITS.
+*> 16.0     THRESHOLD VALUE OF TEST RATIO
+*> 6                 NUMBER OF VALUES OF N
+*> 0 1 2 3 5 9       VALUES OF N
+*> 3                 NUMBER OF VALUES OF ALPHA
+*> 0.0 1.0 0.7       VALUES OF ALPHA
+*> 3                 NUMBER OF VALUES OF BETA
+*> 0.0 1.0 1.3       VALUES OF BETA
+*> DGEMM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> DSYMM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> DTRMM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> DTRSM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> DSYRK  T PUT F FOR NO TEST. SAME COLUMNS.
+*> DSYR2K T PUT F FOR NO TEST. SAME COLUMNS.
+*>
+*> Further Details
+*> ===============
+*>
+*> See:
+*>
+*>    Dongarra J. J., Du Croz J. J., Duff I. S. and Hammarling S.
+*>    A Set of Level 3 Basic Linear Algebra Subprograms.
+*>
+*>    Technical Memorandum No.88 (Revision 1), Mathematics and
+*>    Computer Science Division, Argonne National Laboratory, 9700
+*>    South Cass Avenue, Argonne, Illinois 60439, US.
+*>
+*> -- Written on 8-February-1989.
+*>    Jack Dongarra, Argonne National Laboratory.
+*>    Iain Duff, AERE Harwell.
+*>    Jeremy Du Croz, Numerical Algorithms Group Ltd.
+*>    Sven Hammarling, Numerical Algorithms Group Ltd.
+*>
+*>    10-9-00:  Change STATUS='NEW' to 'UNKNOWN' so that the testers
+*>              can be run multiple times without deleting generated
+*>              output files (susan)
+*> \endverbatim
+*
+*  Authors:
+*  ========
+*
+*> \author Univ. of Tennessee
+*> \author Univ. of California Berkeley
+*> \author Univ. of Colorado Denver
+*> \author NAG Ltd.
+*
+*> \date April 2012
+*
+*> \ingroup double_blas_testing
+*
+*  =====================================================================
       PROGRAM DBLAT3
 *
-*  Test program for the DOUBLE PRECISION Level 3 Blas.
+*  -- Reference BLAS test routine (version 3.7.0) --
+*  -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
+*  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
+*     April 2012
 *
-*  The program must be driven by a short data file. The first 14 records
-*  of the file are read using list-directed input, the last 6 records
-*  are read using the format ( A6, L2 ). An annotated example of a data
-*  file can be obtained by deleting the first 3 characters from the
-*  following 20 lines:
-*  'DBLAT3.SUMM'     NAME OF SUMMARY OUTPUT FILE
-*  6                 UNIT NUMBER OF SUMMARY FILE
-*  'DBLAT3.SNAP'     NAME OF SNAPSHOT OUTPUT FILE
-*  -1                UNIT NUMBER OF SNAPSHOT FILE (NOT USED IF .LT. 0)
-*  F        LOGICAL FLAG, T TO REWIND SNAPSHOT FILE AFTER EACH RECORD.
-*  F        LOGICAL FLAG, T TO STOP ON FAILURES.
-*  T        LOGICAL FLAG, T TO TEST ERROR EXITS.
-*  16.0     THRESHOLD VALUE OF TEST RATIO
-*  6                 NUMBER OF VALUES OF N
-*  0 1 2 3 5 9       VALUES OF N
-*  3                 NUMBER OF VALUES OF ALPHA
-*  0.0 1.0 0.7       VALUES OF ALPHA
-*  3                 NUMBER OF VALUES OF BETA
-*  0.0 1.0 1.3       VALUES OF BETA
-*  DGEMM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  DSYMM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  DTRMM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  DTRSM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  DSYRK  T PUT F FOR NO TEST. SAME COLUMNS.
-*  DSYR2K T PUT F FOR NO TEST. SAME COLUMNS.
-*
-*  See:
-*
-*     Dongarra J. J., Du Croz J. J., Duff I. S. and Hammarling S.
-*     A Set of Level 3 Basic Linear Algebra Subprograms.
-*
-*     Technical Memorandum No.88 (Revision 1), Mathematics and
-*     Computer Science Division, Argonne National Laboratory, 9700
-*     South Cass Avenue, Argonne, Illinois 60439, US.
-*
-*  -- Written on 8-February-1989.
-*     Jack Dongarra, Argonne National Laboratory.
-*     Iain Duff, AERE Harwell.
-*     Jeremy Du Croz, Numerical Algorithms Group Ltd.
-*     Sven Hammarling, Numerical Algorithms Group Ltd.
+*  =====================================================================
 *
 *     .. Parameters ..
       INTEGER            NIN
       PARAMETER          ( NIN = 5 )
       INTEGER            NSUBS
       PARAMETER          ( NSUBS = 6 )
-      DOUBLE PRECISION   ZERO, HALF, ONE
-      PARAMETER          ( ZERO = 0.0D0, HALF = 0.5D0, ONE = 1.0D0 )
+      DOUBLE PRECISION   ZERO, ONE
+      PARAMETER          ( ZERO = 0.0D0, ONE = 1.0D0 )
       INTEGER            NMAX
       PARAMETER          ( NMAX = 65 )
       INTEGER            NIDMAX, NALMAX, NBEMAX
@@ -96,7 +142,7 @@
 *
       READ( NIN, FMT = * )SUMMRY
       READ( NIN, FMT = * )NOUT
-      OPEN( NOUT, FILE = SUMMRY, STATUS = 'NEW' )
+      OPEN( NOUT, FILE = SUMMRY, STATUS = 'UNKNOWN' )
       NOUTC = NOUT
 *
 *     Read name and unit number for snapshot output file and open file.
@@ -105,7 +151,7 @@
       READ( NIN, FMT = * )NTRA
       TRACE = NTRA.GE.0
       IF( TRACE )THEN
-         OPEN( NTRA, FILE = SNAPS, STATUS = 'NEW' )
+         OPEN( NTRA, FILE = SNAPS, STATUS = 'UNKNOWN' )
       END IF
 *     Read the flag that directs rewinding of the snapshot file.
       READ( NIN, FMT = * )REWI
@@ -182,14 +228,7 @@
 *
 *     Compute EPS (the machine precision).
 *
-      EPS = ONE
-   70 CONTINUE
-      IF( DDIFF( ONE + EPS, ONE ).EQ.ZERO )
-     $   GO TO 80
-      EPS = HALF*EPS
-      GO TO 70
-   80 CONTINUE
-      EPS = EPS + EPS
+      EPS = EPSILON(ZERO)
       WRITE( NOUT, FMT = 9998 )EPS
 *
 *     Check the reliability of DMMCH using exact data.
@@ -1802,7 +1841,7 @@
 *
 *  Tests the error exits from the Level 3 Blas.
 *  Requires a special version of the error-handling routine XERBLA.
-*  ALPHA, BETA, A, B and C should not need to be defined.
+*  A, B and C should not need to be defined.
 *
 *  Auxiliary routine for test program for Level 3 Blas.
 *
@@ -1812,12 +1851,18 @@
 *     Jeremy Du Croz, Numerical Algorithms Group Ltd.
 *     Sven Hammarling, Numerical Algorithms Group Ltd.
 *
+*  3-19-92:  Initialize ALPHA and BETA  (eca)
+*  3-19-92:  Fix argument 12 in calls to SSYMM with INFOT = 9  (eca)
+*
 *     .. Scalar Arguments ..
       INTEGER            ISNUM, NOUT
       CHARACTER*6        SRNAMT
 *     .. Scalars in Common ..
       INTEGER            INFOT, NOUTC
       LOGICAL            LERR, OK
+*     .. Parameters ..
+      DOUBLE PRECISION   ONE, TWO
+      PARAMETER          ( ONE = 1.0D0, TWO = 2.0D0 )
 *     .. Local Scalars ..
       DOUBLE PRECISION   ALPHA, BETA
 *     .. Local Arrays ..
@@ -1834,6 +1879,12 @@
 *     LERR is set to .TRUE. by the special version of XERBLA each time
 *     it is called, and is then tested and re-set by CHKXER.
       LERR = .FALSE.
+*
+*     Initialize ALPHA and BETA.
+*
+      ALPHA = ONE
+      BETA = TWO
+*
       GO TO ( 10, 20, 30, 40, 50, 60 )ISNUM
    10 INFOT = 1
       CALL DGEMM( '/', 'N', 0, 0, 0, ALPHA, A, 1, B, 1, BETA, C, 1 )
@@ -1963,16 +2014,16 @@
       CALL DSYMM( 'R', 'L', 0, 2, ALPHA, A, 1, B, 1, BETA, C, 1 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL DSYMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 1 )
+      CALL DSYMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL DSYMM( 'R', 'U', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 1 )
+      CALL DSYMM( 'R', 'U', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL DSYMM( 'L', 'L', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 1 )
+      CALL DSYMM( 'L', 'L', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL DSYMM( 'R', 'L', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 1 )
+      CALL DSYMM( 'R', 'L', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 12
       CALL DSYMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 2, BETA, C, 1 )
@@ -2660,7 +2711,6 @@
    50    CONTINUE
       END IF
 *
-   60 CONTINUE
       LDERES = .TRUE.
       GO TO 80
    70 CONTINUE
diff --git a/test/sblat2.f b/test/sblat2.f
index a1074be52..56ead8640 100644
--- a/test/sblat2.f
+++ b/test/sblat2.f
@@ -1,75 +1,121 @@
+*> \brief \b SBLAT2
+*
+*  =========== DOCUMENTATION ===========
+*
+* Online html documentation available at
+*            http://www.netlib.org/lapack/explore-html/
+*
+*  Definition:
+*  ===========
+*
+*       PROGRAM SBLAT2
+*
+*
+*> \par Purpose:
+*  =============
+*>
+*> \verbatim
+*>
+*> Test program for the REAL Level 2 Blas.
+*>
+*> The program must be driven by a short data file. The first 18 records
+*> of the file are read using list-directed input, the last 16 records
+*> are read using the format ( A6, L2 ). An annotated example of a data
+*> file can be obtained by deleting the first 3 characters from the
+*> following 34 lines:
+*> 'sblat2.out'      NAME OF SUMMARY OUTPUT FILE
+*> 6                 UNIT NUMBER OF SUMMARY FILE
+*> 'SBLAT2.SNAP'     NAME OF SNAPSHOT OUTPUT FILE
+*> -1                UNIT NUMBER OF SNAPSHOT FILE (NOT USED IF .LT. 0)
+*> F        LOGICAL FLAG, T TO REWIND SNAPSHOT FILE AFTER EACH RECORD.
+*> F        LOGICAL FLAG, T TO STOP ON FAILURES.
+*> T        LOGICAL FLAG, T TO TEST ERROR EXITS.
+*> 16.0     THRESHOLD VALUE OF TEST RATIO
+*> 6                 NUMBER OF VALUES OF N
+*> 0 1 2 3 5 9       VALUES OF N
+*> 4                 NUMBER OF VALUES OF K
+*> 0 1 2 4           VALUES OF K
+*> 4                 NUMBER OF VALUES OF INCX AND INCY
+*> 1 2 -1 -2         VALUES OF INCX AND INCY
+*> 3                 NUMBER OF VALUES OF ALPHA
+*> 0.0 1.0 0.7       VALUES OF ALPHA
+*> 3                 NUMBER OF VALUES OF BETA
+*> 0.0 1.0 0.9       VALUES OF BETA
+*> SGEMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> SGBMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> SSYMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> SSBMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> SSPMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> STRMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> STBMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> STPMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> STRSV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> STBSV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> STPSV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> SGER   T PUT F FOR NO TEST. SAME COLUMNS.
+*> SSYR   T PUT F FOR NO TEST. SAME COLUMNS.
+*> SSPR   T PUT F FOR NO TEST. SAME COLUMNS.
+*> SSYR2  T PUT F FOR NO TEST. SAME COLUMNS.
+*> SSPR2  T PUT F FOR NO TEST. SAME COLUMNS.
+*>
+*> Further Details
+*> ===============
+*>
+*>    See:
+*>
+*>       Dongarra J. J., Du Croz J. J., Hammarling S.  and Hanson R. J..
+*>       An  extended  set of Fortran  Basic Linear Algebra Subprograms.
+*>
+*>       Technical  Memoranda  Nos. 41 (revision 3) and 81,  Mathematics
+*>       and  Computer Science  Division,  Argonne  National Laboratory,
+*>       9700 South Cass Avenue, Argonne, Illinois 60439, US.
+*>
+*>       Or
+*>
+*>       NAG  Technical Reports TR3/87 and TR4/87,  Numerical Algorithms
+*>       Group  Ltd.,  NAG  Central  Office,  256  Banbury  Road, Oxford
+*>       OX2 7DE, UK,  and  Numerical Algorithms Group Inc.,  1101  31st
+*>       Street,  Suite 100,  Downers Grove,  Illinois 60515-1263,  USA.
+*>
+*>
+*> -- Written on 10-August-1987.
+*>    Richard Hanson, Sandia National Labs.
+*>    Jeremy Du Croz, NAG Central Office.
+*>
+*>    10-9-00:  Change STATUS='NEW' to 'UNKNOWN' so that the testers
+*>              can be run multiple times without deleting generated
+*>              output files (susan)
+*> \endverbatim
+*
+*  Authors:
+*  ========
+*
+*> \author Univ. of Tennessee
+*> \author Univ. of California Berkeley
+*> \author Univ. of Colorado Denver
+*> \author NAG Ltd.
+*
+*> \date April 2012
+*
+*> \ingroup single_blas_testing
+*
+*  =====================================================================
       PROGRAM SBLAT2
 *
-*  Test program for the REAL             Level 2 Blas.
+*  -- Reference BLAS test routine (version 3.7.0) --
+*  -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
+*  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
+*     April 2012
 *
-*  The program must be driven by a short data file. The first 18 records
-*  of the file are read using list-directed input, the last 16 records
-*  are read using the format ( A6, L2 ). An annotated example of a data
-*  file can be obtained by deleting the first 3 characters from the
-*  following 34 lines:
-*  'SBLAT2.SUMM'     NAME OF SUMMARY OUTPUT FILE
-*  6                 UNIT NUMBER OF SUMMARY FILE
-*  'SBLAT2.SNAP'     NAME OF SNAPSHOT OUTPUT FILE
-*  -1                UNIT NUMBER OF SNAPSHOT FILE (NOT USED IF .LT. 0)
-*  F        LOGICAL FLAG, T TO REWIND SNAPSHOT FILE AFTER EACH RECORD.
-*  F        LOGICAL FLAG, T TO STOP ON FAILURES.
-*  T        LOGICAL FLAG, T TO TEST ERROR EXITS.
-*  16.0     THRESHOLD VALUE OF TEST RATIO
-*  6                 NUMBER OF VALUES OF N
-*  0 1 2 3 5 9       VALUES OF N
-*  4                 NUMBER OF VALUES OF K
-*  0 1 2 4           VALUES OF K
-*  4                 NUMBER OF VALUES OF INCX AND INCY
-*  1 2 -1 -2         VALUES OF INCX AND INCY
-*  3                 NUMBER OF VALUES OF ALPHA
-*  0.0 1.0 0.7       VALUES OF ALPHA
-*  3                 NUMBER OF VALUES OF BETA
-*  0.0 1.0 0.9       VALUES OF BETA
-*  SGEMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  SGBMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  SSYMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  SSBMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  SSPMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  STRMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  STBMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  STPMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  STRSV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  STBSV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  STPSV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  SGER   T PUT F FOR NO TEST. SAME COLUMNS.
-*  SSYR   T PUT F FOR NO TEST. SAME COLUMNS.
-*  SSPR   T PUT F FOR NO TEST. SAME COLUMNS.
-*  SSYR2  T PUT F FOR NO TEST. SAME COLUMNS.
-*  SSPR2  T PUT F FOR NO TEST. SAME COLUMNS.
-*
-*     See:
-*
-*        Dongarra J. J., Du Croz J. J., Hammarling S.  and Hanson R. J..
-*        An  extended  set of Fortran  Basic Linear Algebra Subprograms.
-*
-*        Technical  Memoranda  Nos. 41 (revision 3) and 81,  Mathematics
-*        and  Computer Science  Division,  Argonne  National Laboratory,
-*        9700 South Cass Avenue, Argonne, Illinois 60439, US.
-*
-*        Or
-*
-*        NAG  Technical Reports TR3/87 and TR4/87,  Numerical Algorithms
-*        Group  Ltd.,  NAG  Central  Office,  256  Banbury  Road, Oxford
-*        OX2 7DE, UK,  and  Numerical Algorithms Group Inc.,  1101  31st
-*        Street,  Suite 100,  Downers Grove,  Illinois 60515-1263,  USA.
-*
-*
-*  -- Written on 10-August-1987.
-*     Richard Hanson, Sandia National Labs.
-*     Jeremy Du Croz, NAG Central Office.
+*  =====================================================================
 *
 *     .. Parameters ..
       INTEGER            NIN
       PARAMETER          ( NIN = 5 )
       INTEGER            NSUBS
       PARAMETER          ( NSUBS = 16 )
-      REAL               ZERO, HALF, ONE
-      PARAMETER          ( ZERO = 0.0, HALF = 0.5, ONE = 1.0 )
+      REAL               ZERO, ONE
+      PARAMETER          ( ZERO = 0.0, ONE = 1.0 )
       INTEGER            NMAX, INCMAX
       PARAMETER          ( NMAX = 65, INCMAX = 2 )
       INTEGER            NINMAX, NIDMAX, NKBMAX, NALMAX, NBEMAX
@@ -121,7 +167,7 @@
 *
       READ( NIN, FMT = * )SUMMRY
       READ( NIN, FMT = * )NOUT
-      OPEN( NOUT, FILE = SUMMRY, STATUS = 'NEW' )
+      OPEN( NOUT, FILE = SUMMRY, STATUS = 'UNKNOWN' )
       NOUTC = NOUT
 *
 *     Read name and unit number for snapshot output file and open file.
@@ -130,7 +176,7 @@
       READ( NIN, FMT = * )NTRA
       TRACE = NTRA.GE.0
       IF( TRACE )THEN
-         OPEN( NTRA, FILE = SNAPS, STATUS = 'NEW' )
+         OPEN( NTRA, FILE = SNAPS, STATUS = 'UNKNOWN' )
       END IF
 *     Read the flag that directs rewinding of the snapshot file.
       READ( NIN, FMT = * )REWI
@@ -235,14 +281,7 @@
 *
 *     Compute EPS (the machine precision).
 *
-      EPS = ONE
-   90 CONTINUE
-      IF( SDIFF( ONE + EPS, ONE ).EQ.ZERO )
-     $   GO TO 100
-      EPS = HALF*EPS
-      GO TO 90
-  100 CONTINUE
-      EPS = EPS + EPS
+      EPS = EPSILON(ZERO)
       WRITE( NOUT, FMT = 9998 )EPS
 *
 *     Check the reliability of SMVCH using exact data.
@@ -2982,7 +3021,6 @@
    50    CONTINUE
       END IF
 *
-   60 CONTINUE
       LSERES = .TRUE.
       GO TO 80
    70 CONTINUE
diff --git a/test/sblat3.f b/test/sblat3.f
index 325a9eb92..66edac14e 100644
--- a/test/sblat3.f
+++ b/test/sblat3.f
@@ -1,55 +1,101 @@
+*> \brief \b SBLAT3
+*
+*  =========== DOCUMENTATION ===========
+*
+* Online html documentation available at
+*            http://www.netlib.org/lapack/explore-html/
+*
+*  Definition:
+*  ===========
+*
+*       PROGRAM SBLAT3
+*
+*
+*> \par Purpose:
+*  =============
+*>
+*> \verbatim
+*>
+*> Test program for the REAL             Level 3 Blas.
+*>
+*> The program must be driven by a short data file. The first 14 records
+*> of the file are read using list-directed input, the last 6 records
+*> are read using the format ( A6, L2 ). An annotated example of a data
+*> file can be obtained by deleting the first 3 characters from the
+*> following 20 lines:
+*> 'sblat3.out'      NAME OF SUMMARY OUTPUT FILE
+*> 6                 UNIT NUMBER OF SUMMARY FILE
+*> 'SBLAT3.SNAP'     NAME OF SNAPSHOT OUTPUT FILE
+*> -1                UNIT NUMBER OF SNAPSHOT FILE (NOT USED IF .LT. 0)
+*> F        LOGICAL FLAG, T TO REWIND SNAPSHOT FILE AFTER EACH RECORD.
+*> F        LOGICAL FLAG, T TO STOP ON FAILURES.
+*> T        LOGICAL FLAG, T TO TEST ERROR EXITS.
+*> 16.0     THRESHOLD VALUE OF TEST RATIO
+*> 6                 NUMBER OF VALUES OF N
+*> 0 1 2 3 5 9       VALUES OF N
+*> 3                 NUMBER OF VALUES OF ALPHA
+*> 0.0 1.0 0.7       VALUES OF ALPHA
+*> 3                 NUMBER OF VALUES OF BETA
+*> 0.0 1.0 1.3       VALUES OF BETA
+*> SGEMM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> SSYMM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> STRMM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> STRSM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> SSYRK  T PUT F FOR NO TEST. SAME COLUMNS.
+*> SSYR2K T PUT F FOR NO TEST. SAME COLUMNS.
+*>
+*> Further Details
+*> ===============
+*>
+*> See:
+*>
+*>    Dongarra J. J., Du Croz J. J., Duff I. S. and Hammarling S.
+*>    A Set of Level 3 Basic Linear Algebra Subprograms.
+*>
+*>    Technical Memorandum No.88 (Revision 1), Mathematics and
+*>    Computer Science Division, Argonne National Laboratory, 9700
+*>    South Cass Avenue, Argonne, Illinois 60439, US.
+*>
+*> -- Written on 8-February-1989.
+*>    Jack Dongarra, Argonne National Laboratory.
+*>    Iain Duff, AERE Harwell.
+*>    Jeremy Du Croz, Numerical Algorithms Group Ltd.
+*>    Sven Hammarling, Numerical Algorithms Group Ltd.
+*>
+*>    10-9-00:  Change STATUS='NEW' to 'UNKNOWN' so that the testers
+*>              can be run multiple times without deleting generated
+*>              output files (susan)
+*> \endverbatim
+*
+*  Authors:
+*  ========
+*
+*> \author Univ. of Tennessee
+*> \author Univ. of California Berkeley
+*> \author Univ. of Colorado Denver
+*> \author NAG Ltd.
+*
+*> \date April 2012
+*
+*> \ingroup single_blas_testing
+*
+*  =====================================================================
       PROGRAM SBLAT3
 *
-*  Test program for the REAL             Level 3 Blas.
+*  -- Reference BLAS test routine (version 3.7.0) --
+*  -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
+*  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
+*     April 2012
 *
-*  The program must be driven by a short data file. The first 14 records
-*  of the file are read using list-directed input, the last 6 records
-*  are read using the format ( A6, L2 ). An annotated example of a data
-*  file can be obtained by deleting the first 3 characters from the
-*  following 20 lines:
-*  'SBLAT3.SUMM'     NAME OF SUMMARY OUTPUT FILE
-*  6                 UNIT NUMBER OF SUMMARY FILE
-*  'SBLAT3.SNAP'     NAME OF SNAPSHOT OUTPUT FILE
-*  -1                UNIT NUMBER OF SNAPSHOT FILE (NOT USED IF .LT. 0)
-*  F        LOGICAL FLAG, T TO REWIND SNAPSHOT FILE AFTER EACH RECORD.
-*  F        LOGICAL FLAG, T TO STOP ON FAILURES.
-*  T        LOGICAL FLAG, T TO TEST ERROR EXITS.
-*  16.0     THRESHOLD VALUE OF TEST RATIO
-*  6                 NUMBER OF VALUES OF N
-*  0 1 2 3 5 9       VALUES OF N
-*  3                 NUMBER OF VALUES OF ALPHA
-*  0.0 1.0 0.7       VALUES OF ALPHA
-*  3                 NUMBER OF VALUES OF BETA
-*  0.0 1.0 1.3       VALUES OF BETA
-*  SGEMM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  SSYMM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  STRMM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  STRSM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  SSYRK  T PUT F FOR NO TEST. SAME COLUMNS.
-*  SSYR2K T PUT F FOR NO TEST. SAME COLUMNS.
-*
-*  See:
-*
-*     Dongarra J. J., Du Croz J. J., Duff I. S. and Hammarling S.
-*     A Set of Level 3 Basic Linear Algebra Subprograms.
-*
-*     Technical Memorandum No.88 (Revision 1), Mathematics and
-*     Computer Science Division, Argonne National Laboratory, 9700
-*     South Cass Avenue, Argonne, Illinois 60439, US.
-*
-*  -- Written on 8-February-1989.
-*     Jack Dongarra, Argonne National Laboratory.
-*     Iain Duff, AERE Harwell.
-*     Jeremy Du Croz, Numerical Algorithms Group Ltd.
-*     Sven Hammarling, Numerical Algorithms Group Ltd.
+*  =====================================================================
 *
 *     .. Parameters ..
       INTEGER            NIN
       PARAMETER          ( NIN = 5 )
       INTEGER            NSUBS
       PARAMETER          ( NSUBS = 6 )
-      REAL               ZERO, HALF, ONE
-      PARAMETER          ( ZERO = 0.0, HALF = 0.5, ONE = 1.0 )
+      REAL               ZERO, ONE
+      PARAMETER          ( ZERO = 0.0, ONE = 1.0 )
       INTEGER            NMAX
       PARAMETER          ( NMAX = 65 )
       INTEGER            NIDMAX, NALMAX, NBEMAX
@@ -96,7 +142,7 @@
 *
       READ( NIN, FMT = * )SUMMRY
       READ( NIN, FMT = * )NOUT
-      OPEN( NOUT, FILE = SUMMRY, STATUS = 'NEW' )
+      OPEN( NOUT, FILE = SUMMRY )
       NOUTC = NOUT
 *
 *     Read name and unit number for snapshot output file and open file.
@@ -105,7 +151,7 @@
       READ( NIN, FMT = * )NTRA
       TRACE = NTRA.GE.0
       IF( TRACE )THEN
-         OPEN( NTRA, FILE = SNAPS, STATUS = 'NEW' )
+         OPEN( NTRA, FILE = SNAPS )
       END IF
 *     Read the flag that directs rewinding of the snapshot file.
       READ( NIN, FMT = * )REWI
@@ -182,14 +228,7 @@
 *
 *     Compute EPS (the machine precision).
 *
-      EPS = ONE
-   70 CONTINUE
-      IF( SDIFF( ONE + EPS, ONE ).EQ.ZERO )
-     $   GO TO 80
-      EPS = HALF*EPS
-      GO TO 70
-   80 CONTINUE
-      EPS = EPS + EPS
+      EPS = EPSILON(ZERO)
       WRITE( NOUT, FMT = 9998 )EPS
 *
 *     Check the reliability of SMMCH using exact data.
@@ -1802,7 +1841,7 @@
 *
 *  Tests the error exits from the Level 3 Blas.
 *  Requires a special version of the error-handling routine XERBLA.
-*  ALPHA, BETA, A, B and C should not need to be defined.
+*  A, B and C should not need to be defined.
 *
 *  Auxiliary routine for test program for Level 3 Blas.
 *
@@ -1812,12 +1851,18 @@
 *     Jeremy Du Croz, Numerical Algorithms Group Ltd.
 *     Sven Hammarling, Numerical Algorithms Group Ltd.
 *
+*  3-19-92:  Initialize ALPHA and BETA  (eca)
+*  3-19-92:  Fix argument 12 in calls to SSYMM with INFOT = 9  (eca)
+*
 *     .. Scalar Arguments ..
       INTEGER            ISNUM, NOUT
       CHARACTER*6        SRNAMT
 *     .. Scalars in Common ..
       INTEGER            INFOT, NOUTC
       LOGICAL            LERR, OK
+*     .. Parameters ..
+      REAL               ONE, TWO
+      PARAMETER          ( ONE = 1.0E0, TWO = 2.0E0 )
 *     .. Local Scalars ..
       REAL               ALPHA, BETA
 *     .. Local Arrays ..
@@ -1834,6 +1879,12 @@
 *     LERR is set to .TRUE. by the special version of XERBLA each time
 *     it is called, and is then tested and re-set by CHKXER.
       LERR = .FALSE.
+*
+*     Initialize ALPHA and BETA.
+*
+      ALPHA = ONE
+      BETA = TWO
+*
       GO TO ( 10, 20, 30, 40, 50, 60 )ISNUM
    10 INFOT = 1
       CALL SGEMM( '/', 'N', 0, 0, 0, ALPHA, A, 1, B, 1, BETA, C, 1 )
@@ -1963,16 +2014,16 @@
       CALL SSYMM( 'R', 'L', 0, 2, ALPHA, A, 1, B, 1, BETA, C, 1 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL SSYMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 1 )
+      CALL SSYMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL SSYMM( 'R', 'U', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 1 )
+      CALL SSYMM( 'R', 'U', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL SSYMM( 'L', 'L', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 1 )
+      CALL SSYMM( 'L', 'L', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL SSYMM( 'R', 'L', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 1 )
+      CALL SSYMM( 'R', 'L', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 12
       CALL SSYMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 2, BETA, C, 1 )
@@ -2660,7 +2711,6 @@
    50    CONTINUE
       END IF
 *
-   60 CONTINUE
       LSERES = .TRUE.
       GO TO 80
    70 CONTINUE
diff --git a/test/zblat1.f b/test/zblat1.f
index 8b4b8d21e..2d7b88490 100644
--- a/test/zblat1.f
+++ b/test/zblat1.f
@@ -1,7 +1,49 @@
+*> \brief \b ZBLAT1
+*
+*  =========== DOCUMENTATION ===========
+*
+* Online html documentation available at
+*            http://www.netlib.org/lapack/explore-html/
+*
+*  Definition:
+*  ===========
+*
+*       PROGRAM ZBLAT1
+*
+*
+*> \par Purpose:
+*  =============
+*>
+*> \verbatim
+*>
+*>    Test program for the COMPLEX*16 Level 1 BLAS.
+*>
+*>    Based upon the original BLAS test routine together with:
+*>    F06GAF Example Program Text
+*> \endverbatim
+*
+*  Authors:
+*  ========
+*
+*> \author Univ. of Tennessee
+*> \author Univ. of California Berkeley
+*> \author Univ. of Colorado Denver
+*> \author NAG Ltd.
+*
+*> \date April 2012
+*
+*> \ingroup complex16_blas_testing
+*
+*  =====================================================================
       PROGRAM ZBLAT1
-*     Test program for the COMPLEX*16 Level 1 BLAS.
-*     Based upon the original BLAS test routine together with:
-*     F06GAF Example Program Text
+*
+*  -- Reference BLAS test routine (version 3.7.0) --
+*  -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
+*  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
+*     April 2012
+*
+*  =====================================================================
+*
 *     .. Parameters ..
       INTEGER          NOUT
       PARAMETER        (NOUT=6)
@@ -114,8 +156,8 @@
      +                  (5.0D0,6.0D0), (5.0D0,6.0D0), (0.1D0,0.1D0),
      +                  (-0.6D0,0.1D0), (0.1D0,-0.3D0), (7.0D0,8.0D0),
      +                  (7.0D0,8.0D0), (7.0D0,8.0D0), (7.0D0,8.0D0),
-     +                  (7.0D0,8.0D0), (0.3D0,0.1D0), (0.1D0,0.4D0),
-     +                  (0.4D0,0.1D0), (0.1D0,0.2D0), (2.0D0,3.0D0),
+     +                  (7.0D0,8.0D0), (0.3D0,0.1D0), (0.5D0,0.0D0),
+     +                  (0.0D0,0.5D0), (0.0D0,0.2D0), (2.0D0,3.0D0),
      +                  (2.0D0,3.0D0), (2.0D0,3.0D0), (2.0D0,3.0D0)/
       DATA              ((CV(I,J,2),I=1,8),J=1,5)/(0.1D0,0.1D0),
      +                  (4.0D0,5.0D0), (4.0D0,5.0D0), (4.0D0,5.0D0),
@@ -129,10 +171,10 @@
      +                  (3.0D0,6.0D0), (-0.6D0,0.1D0), (4.0D0,7.0D0),
      +                  (0.1D0,-0.3D0), (7.0D0,2.0D0), (7.0D0,2.0D0),
      +                  (7.0D0,2.0D0), (0.3D0,0.1D0), (5.0D0,8.0D0),
-     +                  (0.1D0,0.4D0), (6.0D0,9.0D0), (0.4D0,0.1D0),
-     +                  (8.0D0,3.0D0), (0.1D0,0.2D0), (9.0D0,4.0D0)/
-      DATA              STRUE2/0.0D0, 0.5D0, 0.6D0, 0.7D0, 0.7D0/
-      DATA              STRUE4/0.0D0, 0.7D0, 1.0D0, 1.3D0, 1.7D0/
+     +                  (0.5D0,0.0D0), (6.0D0,9.0D0), (0.0D0,0.5D0),
+     +                  (8.0D0,3.0D0), (0.0D0,0.2D0), (9.0D0,4.0D0)/
+      DATA              STRUE2/0.0D0, 0.5D0, 0.6D0, 0.7D0, 0.8D0/
+      DATA              STRUE4/0.0D0, 0.7D0, 1.0D0, 1.3D0, 1.6D0/
       DATA              ((CTRUE5(I,J,1),I=1,8),J=1,5)/(0.1D0,0.1D0),
      +                  (1.0D0,2.0D0), (1.0D0,2.0D0), (1.0D0,2.0D0),
      +                  (1.0D0,2.0D0), (1.0D0,2.0D0), (1.0D0,2.0D0),
@@ -145,8 +187,8 @@
      +                  (0.11D0,-0.03D0), (-0.17D0,0.46D0),
      +                  (-0.17D0,-0.19D0), (7.0D0,8.0D0), (7.0D0,8.0D0),
      +                  (7.0D0,8.0D0), (7.0D0,8.0D0), (7.0D0,8.0D0),
-     +                  (0.19D0,-0.17D0), (0.32D0,0.09D0),
-     +                  (0.23D0,-0.24D0), (0.18D0,0.01D0),
+     +                  (0.19D0,-0.17D0), (0.20D0,-0.35D0),
+     +                  (0.35D0,0.20D0), (0.14D0,0.08D0),
      +                  (2.0D0,3.0D0), (2.0D0,3.0D0), (2.0D0,3.0D0),
      +                  (2.0D0,3.0D0)/
       DATA              ((CTRUE5(I,J,2),I=1,8),J=1,5)/(0.1D0,0.1D0),
@@ -162,9 +204,9 @@
      +                  (-0.17D0,0.46D0), (4.0D0,7.0D0),
      +                  (-0.17D0,-0.19D0), (7.0D0,2.0D0), (7.0D0,2.0D0),
      +                  (7.0D0,2.0D0), (0.19D0,-0.17D0), (5.0D0,8.0D0),
-     +                  (0.32D0,0.09D0), (6.0D0,9.0D0),
-     +                  (0.23D0,-0.24D0), (8.0D0,3.0D0),
-     +                  (0.18D0,0.01D0), (9.0D0,4.0D0)/
+     +                  (0.20D0,-0.35D0), (6.0D0,9.0D0),
+     +                  (0.35D0,0.20D0), (8.0D0,3.0D0),
+     +                  (0.14D0,0.08D0), (9.0D0,4.0D0)/
       DATA              ((CTRUE6(I,J,1),I=1,8),J=1,5)/(0.1D0,0.1D0),
      +                  (1.0D0,2.0D0), (1.0D0,2.0D0), (1.0D0,2.0D0),
      +                  (1.0D0,2.0D0), (1.0D0,2.0D0), (1.0D0,2.0D0),
@@ -177,8 +219,8 @@
      +                  (0.03D0,0.03D0), (-0.18D0,0.03D0),
      +                  (0.03D0,-0.09D0), (7.0D0,8.0D0), (7.0D0,8.0D0),
      +                  (7.0D0,8.0D0), (7.0D0,8.0D0), (7.0D0,8.0D0),
-     +                  (0.09D0,0.03D0), (0.03D0,0.12D0),
-     +                  (0.12D0,0.03D0), (0.03D0,0.06D0), (2.0D0,3.0D0),
+     +                  (0.09D0,0.03D0), (0.15D0,0.00D0),
+     +                  (0.00D0,0.15D0), (0.00D0,0.06D0), (2.0D0,3.0D0),
      +                  (2.0D0,3.0D0), (2.0D0,3.0D0), (2.0D0,3.0D0)/
       DATA              ((CTRUE6(I,J,2),I=1,8),J=1,5)/(0.1D0,0.1D0),
      +                  (4.0D0,5.0D0), (4.0D0,5.0D0), (4.0D0,5.0D0),
@@ -193,8 +235,8 @@
      +                  (-0.18D0,0.03D0), (4.0D0,7.0D0),
      +                  (0.03D0,-0.09D0), (7.0D0,2.0D0), (7.0D0,2.0D0),
      +                  (7.0D0,2.0D0), (0.09D0,0.03D0), (5.0D0,8.0D0),
-     +                  (0.03D0,0.12D0), (6.0D0,9.0D0), (0.12D0,0.03D0),
-     +                  (8.0D0,3.0D0), (0.03D0,0.06D0), (9.0D0,4.0D0)/
+     +                  (0.15D0,0.00D0), (6.0D0,9.0D0), (0.00D0,0.15D0),
+     +                  (8.0D0,3.0D0), (0.00D0,0.06D0), (9.0D0,4.0D0)/
       DATA              ITRUE3/0, 1, 2, 2, 2/
 *     .. Executable Statements ..
       DO 60 INCX = 1, 2
@@ -529,7 +571,8 @@
 *
 *     .. Parameters ..
       INTEGER          NOUT
-      PARAMETER        (NOUT=6)
+      DOUBLE PRECISION ZERO
+      PARAMETER        (NOUT=6, ZERO=0.0D0)
 *     .. Scalar Arguments ..
       DOUBLE PRECISION SFAC
       INTEGER          LEN
@@ -552,7 +595,7 @@
 *
       DO 40 I = 1, LEN
          SD = SCOMP(I) - STRUE(I)
-         IF (SDIFF(ABS(SSIZE(I))+ABS(SFAC*SD),ABS(SSIZE(I))).EQ.0.0D0)
+         IF (ABS(SFAC*SD) .LE. ABS(SSIZE(I))*EPSILON(ZERO))
      +       GO TO 40
 *
 *                             HERE    SCOMP(I) IS NOT CLOSE TO STRUE(I).
diff --git a/test/zblat2.f b/test/zblat2.f
index e65cdcc70..4a20ac567 100644
--- a/test/zblat2.f
+++ b/test/zblat2.f
@@ -1,68 +1,114 @@
+*> \brief \b ZBLAT2
+*
+*  =========== DOCUMENTATION ===========
+*
+* Online html documentation available at
+*            http://www.netlib.org/lapack/explore-html/
+*
+*  Definition:
+*  ===========
+*
+*       PROGRAM ZBLAT2
+*
+*
+*> \par Purpose:
+*  =============
+*>
+*> \verbatim
+*>
+*> Test program for the COMPLEX*16       Level 2 Blas.
+*>
+*> The program must be driven by a short data file. The first 18 records
+*> of the file are read using list-directed input, the last 17 records
+*> are read using the format ( A6, L2 ). An annotated example of a data
+*> file can be obtained by deleting the first 3 characters from the
+*> following 35 lines:
+*> 'zblat2.out'      NAME OF SUMMARY OUTPUT FILE
+*> 6                 UNIT NUMBER OF SUMMARY FILE
+*> 'CBLA2T.SNAP'     NAME OF SNAPSHOT OUTPUT FILE
+*> -1                UNIT NUMBER OF SNAPSHOT FILE (NOT USED IF .LT. 0)
+*> F        LOGICAL FLAG, T TO REWIND SNAPSHOT FILE AFTER EACH RECORD.
+*> F        LOGICAL FLAG, T TO STOP ON FAILURES.
+*> T        LOGICAL FLAG, T TO TEST ERROR EXITS.
+*> 16.0     THRESHOLD VALUE OF TEST RATIO
+*> 6                 NUMBER OF VALUES OF N
+*> 0 1 2 3 5 9       VALUES OF N
+*> 4                 NUMBER OF VALUES OF K
+*> 0 1 2 4           VALUES OF K
+*> 4                 NUMBER OF VALUES OF INCX AND INCY
+*> 1 2 -1 -2         VALUES OF INCX AND INCY
+*> 3                 NUMBER OF VALUES OF ALPHA
+*> (0.0,0.0) (1.0,0.0) (0.7,-0.9)       VALUES OF ALPHA
+*> 3                 NUMBER OF VALUES OF BETA
+*> (0.0,0.0) (1.0,0.0) (1.3,-1.1)       VALUES OF BETA
+*> ZGEMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZGBMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZHEMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZHBMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZHPMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZTRMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZTBMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZTPMV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZTRSV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZTBSV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZTPSV  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZGERC  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZGERU  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZHER   T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZHPR   T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZHER2  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZHPR2  T PUT F FOR NO TEST. SAME COLUMNS.
+*>
+*> Further Details
+*> ===============
+*>
+*>    See:
+*>
+*>       Dongarra J. J., Du Croz J. J., Hammarling S.  and Hanson R. J..
+*>       An  extended  set of Fortran  Basic Linear Algebra Subprograms.
+*>
+*>       Technical  Memoranda  Nos. 41 (revision 3) and 81,  Mathematics
+*>       and  Computer Science  Division,  Argonne  National Laboratory,
+*>       9700 South Cass Avenue, Argonne, Illinois 60439, US.
+*>
+*>       Or
+*>
+*>       NAG  Technical Reports TR3/87 and TR4/87,  Numerical Algorithms
+*>       Group  Ltd.,  NAG  Central  Office,  256  Banbury  Road, Oxford
+*>       OX2 7DE, UK,  and  Numerical Algorithms Group Inc.,  1101  31st
+*>       Street,  Suite 100,  Downers Grove,  Illinois 60515-1263,  USA.
+*>
+*>
+*> -- Written on 10-August-1987.
+*>    Richard Hanson, Sandia National Labs.
+*>    Jeremy Du Croz, NAG Central Office.
+*>
+*>    10-9-00:  Change STATUS='NEW' to 'UNKNOWN' so that the testers
+*>              can be run multiple times without deleting generated
+*>              output files (susan)
+*> \endverbatim
+*
+*  Authors:
+*  ========
+*
+*> \author Univ. of Tennessee
+*> \author Univ. of California Berkeley
+*> \author Univ. of Colorado Denver
+*> \author NAG Ltd.
+*
+*> \date April 2012
+*
+*> \ingroup complex16_blas_testing
+*
+*  =====================================================================
       PROGRAM ZBLAT2
 *
-*  Test program for the COMPLEX*16       Level 2 Blas.
+*  -- Reference BLAS test routine (version 3.7.0) --
+*  -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
+*  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
+*     April 2012
 *
-*  The program must be driven by a short data file. The first 18 records
-*  of the file are read using list-directed input, the last 17 records
-*  are read using the format ( A6, L2 ). An annotated example of a data
-*  file can be obtained by deleting the first 3 characters from the
-*  following 35 lines:
-*  'ZBLAT2.SUMM'     NAME OF SUMMARY OUTPUT FILE
-*  6                 UNIT NUMBER OF SUMMARY FILE
-*  'CBLA2T.SNAP'     NAME OF SNAPSHOT OUTPUT FILE
-*  -1                UNIT NUMBER OF SNAPSHOT FILE (NOT USED IF .LT. 0)
-*  F        LOGICAL FLAG, T TO REWIND SNAPSHOT FILE AFTER EACH RECORD.
-*  F        LOGICAL FLAG, T TO STOP ON FAILURES.
-*  T        LOGICAL FLAG, T TO TEST ERROR EXITS.
-*  16.0     THRESHOLD VALUE OF TEST RATIO
-*  6                 NUMBER OF VALUES OF N
-*  0 1 2 3 5 9       VALUES OF N
-*  4                 NUMBER OF VALUES OF K
-*  0 1 2 4           VALUES OF K
-*  4                 NUMBER OF VALUES OF INCX AND INCY
-*  1 2 -1 -2         VALUES OF INCX AND INCY
-*  3                 NUMBER OF VALUES OF ALPHA
-*  (0.0,0.0) (1.0,0.0) (0.7,-0.9)       VALUES OF ALPHA
-*  3                 NUMBER OF VALUES OF BETA
-*  (0.0,0.0) (1.0,0.0) (1.3,-1.1)       VALUES OF BETA
-*  ZGEMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZGBMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZHEMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZHBMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZHPMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZTRMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZTBMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZTPMV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZTRSV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZTBSV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZTPSV  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZGERC  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZGERU  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZHER   T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZHPR   T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZHER2  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZHPR2  T PUT F FOR NO TEST. SAME COLUMNS.
-*
-*     See:
-*
-*        Dongarra J. J., Du Croz J. J., Hammarling S.  and Hanson R. J..
-*        An  extended  set of Fortran  Basic Linear Algebra Subprograms.
-*
-*        Technical  Memoranda  Nos. 41 (revision 3) and 81,  Mathematics
-*        and  Computer Science  Division,  Argonne  National Laboratory,
-*        9700 South Cass Avenue, Argonne, Illinois 60439, US.
-*
-*        Or
-*
-*        NAG  Technical Reports TR3/87 and TR4/87,  Numerical Algorithms
-*        Group  Ltd.,  NAG  Central  Office,  256  Banbury  Road, Oxford
-*        OX2 7DE, UK,  and  Numerical Algorithms Group Inc.,  1101  31st
-*        Street,  Suite 100,  Downers Grove,  Illinois 60515-1263,  USA.
-*
-*
-*  -- Written on 10-August-1987.
-*     Richard Hanson, Sandia National Labs.
-*     Jeremy Du Croz, NAG Central Office.
+*  =====================================================================
 *
 *     .. Parameters ..
       INTEGER            NIN
@@ -72,8 +118,8 @@
       COMPLEX*16         ZERO, ONE
       PARAMETER          ( ZERO = ( 0.0D0, 0.0D0 ),
      $                   ONE = ( 1.0D0, 0.0D0 ) )
-      DOUBLE PRECISION   RZERO, RHALF, RONE
-      PARAMETER          ( RZERO = 0.0D0, RHALF = 0.5D0, RONE = 1.0D0 )
+      DOUBLE PRECISION   RZERO
+      PARAMETER          ( RZERO = 0.0D0 )
       INTEGER            NMAX, INCMAX
       PARAMETER          ( NMAX = 65, INCMAX = 2 )
       INTEGER            NINMAX, NIDMAX, NKBMAX, NALMAX, NBEMAX
@@ -127,7 +173,7 @@
 *
       READ( NIN, FMT = * )SUMMRY
       READ( NIN, FMT = * )NOUT
-      OPEN( NOUT, FILE = SUMMRY, STATUS = 'NEW' )
+      OPEN( NOUT, FILE = SUMMRY, STATUS = 'UNKNOWN' )
       NOUTC = NOUT
 *
 *     Read name and unit number for snapshot output file and open file.
@@ -136,7 +182,7 @@
       READ( NIN, FMT = * )NTRA
       TRACE = NTRA.GE.0
       IF( TRACE )THEN
-         OPEN( NTRA, FILE = SNAPS, STATUS = 'NEW' )
+         OPEN( NTRA, FILE = SNAPS, STATUS = 'UNKNOWN' )
       END IF
 *     Read the flag that directs rewinding of the snapshot file.
       READ( NIN, FMT = * )REWI
@@ -241,14 +287,7 @@
 *
 *     Compute EPS (the machine precision).
 *
-      EPS = RONE
-   90 CONTINUE
-      IF( DDIFF( RONE + EPS, RONE ).EQ.RZERO )
-     $   GO TO 100
-      EPS = RHALF*EPS
-      GO TO 90
-  100 CONTINUE
-      EPS = EPS + EPS
+      EPS = EPSILON(RZERO)
       WRITE( NOUT, FMT = 9998 )EPS
 *
 *     Check the reliability of ZMVCH using exact data.
@@ -3087,7 +3126,6 @@
    50    CONTINUE
       END IF
 *
-   60 CONTINUE
       LZERES = .TRUE.
       GO TO 80
    70 CONTINUE
diff --git a/test/zblat3.f b/test/zblat3.f
index f03b1a617..0e38334e9 100644
--- a/test/zblat3.f
+++ b/test/zblat3.f
@@ -1,50 +1,97 @@
+*> \brief \b ZBLAT3
+*
+*  =========== DOCUMENTATION ===========
+*
+* Online html documentation available at
+*            http://www.netlib.org/lapack/explore-html/
+*
+*  Definition:
+*  ===========
+*
+*       PROGRAM ZBLAT3
+*
+*
+*> \par Purpose:
+*  =============
+*>
+*> \verbatim
+*>
+*> Test program for the COMPLEX*16       Level 3 Blas.
+*>
+*> The program must be driven by a short data file. The first 14 records
+*> of the file are read using list-directed input, the last 9 records
+*> are read using the format ( A6, L2 ). An annotated example of a data
+*> file can be obtained by deleting the first 3 characters from the
+*> following 23 lines:
+*> 'zblat3.out'      NAME OF SUMMARY OUTPUT FILE
+*> 6                 UNIT NUMBER OF SUMMARY FILE
+*> 'ZBLAT3.SNAP'     NAME OF SNAPSHOT OUTPUT FILE
+*> -1                UNIT NUMBER OF SNAPSHOT FILE (NOT USED IF .LT. 0)
+*> F        LOGICAL FLAG, T TO REWIND SNAPSHOT FILE AFTER EACH RECORD.
+*> F        LOGICAL FLAG, T TO STOP ON FAILURES.
+*> T        LOGICAL FLAG, T TO TEST ERROR EXITS.
+*> 16.0     THRESHOLD VALUE OF TEST RATIO
+*> 6                 NUMBER OF VALUES OF N
+*> 0 1 2 3 5 9       VALUES OF N
+*> 3                 NUMBER OF VALUES OF ALPHA
+*> (0.0,0.0) (1.0,0.0) (0.7,-0.9)       VALUES OF ALPHA
+*> 3                 NUMBER OF VALUES OF BETA
+*> (0.0,0.0) (1.0,0.0) (1.3,-1.1)       VALUES OF BETA
+*> ZGEMM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZHEMM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZSYMM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZTRMM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZTRSM  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZHERK  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZSYRK  T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZHER2K T PUT F FOR NO TEST. SAME COLUMNS.
+*> ZSYR2K T PUT F FOR NO TEST. SAME COLUMNS.
+*>
+*>
+*> Further Details
+*> ===============
+*>
+*> See:
+*>
+*>    Dongarra J. J., Du Croz J. J., Duff I. S. and Hammarling S.
+*>    A Set of Level 3 Basic Linear Algebra Subprograms.
+*>
+*>    Technical Memorandum No.88 (Revision 1), Mathematics and
+*>    Computer Science Division, Argonne National Laboratory, 9700
+*>    South Cass Avenue, Argonne, Illinois 60439, US.
+*>
+*> -- Written on 8-February-1989.
+*>    Jack Dongarra, Argonne National Laboratory.
+*>    Iain Duff, AERE Harwell.
+*>    Jeremy Du Croz, Numerical Algorithms Group Ltd.
+*>    Sven Hammarling, Numerical Algorithms Group Ltd.
+*>
+*>    10-9-00:  Change STATUS='NEW' to 'UNKNOWN' so that the testers
+*>              can be run multiple times without deleting generated
+*>              output files (susan)
+*> \endverbatim
+*
+*  Authors:
+*  ========
+*
+*> \author Univ. of Tennessee
+*> \author Univ. of California Berkeley
+*> \author Univ. of Colorado Denver
+*> \author NAG Ltd.
+*
+*> \date April 2012
+*
+*> \ingroup complex16_blas_testing
+*
+*  =====================================================================
       PROGRAM ZBLAT3
 *
-*  Test program for the COMPLEX*16       Level 3 Blas.
+*  -- Reference BLAS test routine (version 3.7.0) --
+*  -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
+*  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
+*     April 2012
 *
-*  The program must be driven by a short data file. The first 14 records
-*  of the file are read using list-directed input, the last 9 records
-*  are read using the format ( A6, L2 ). An annotated example of a data
-*  file can be obtained by deleting the first 3 characters from the
-*  following 23 lines:
-*  'ZBLAT3.SUMM'     NAME OF SUMMARY OUTPUT FILE
-*  6                 UNIT NUMBER OF SUMMARY FILE
-*  'ZBLAT3.SNAP'     NAME OF SNAPSHOT OUTPUT FILE
-*  -1                UNIT NUMBER OF SNAPSHOT FILE (NOT USED IF .LT. 0)
-*  F        LOGICAL FLAG, T TO REWIND SNAPSHOT FILE AFTER EACH RECORD.
-*  F        LOGICAL FLAG, T TO STOP ON FAILURES.
-*  T        LOGICAL FLAG, T TO TEST ERROR EXITS.
-*  16.0     THRESHOLD VALUE OF TEST RATIO
-*  6                 NUMBER OF VALUES OF N
-*  0 1 2 3 5 9       VALUES OF N
-*  3                 NUMBER OF VALUES OF ALPHA
-*  (0.0,0.0) (1.0,0.0) (0.7,-0.9)       VALUES OF ALPHA
-*  3                 NUMBER OF VALUES OF BETA
-*  (0.0,0.0) (1.0,0.0) (1.3,-1.1)       VALUES OF BETA
-*  ZGEMM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZHEMM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZSYMM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZTRMM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZTRSM  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZHERK  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZSYRK  T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZHER2K T PUT F FOR NO TEST. SAME COLUMNS.
-*  ZSYR2K T PUT F FOR NO TEST. SAME COLUMNS.
-*
-*  See:
-*
-*     Dongarra J. J., Du Croz J. J., Duff I. S. and Hammarling S.
-*     A Set of Level 3 Basic Linear Algebra Subprograms.
-*
-*     Technical Memorandum No.88 (Revision 1), Mathematics and
-*     Computer Science Division, Argonne National Laboratory, 9700
-*     South Cass Avenue, Argonne, Illinois 60439, US.
-*
-*  -- Written on 8-February-1989.
-*     Jack Dongarra, Argonne National Laboratory.
-*     Iain Duff, AERE Harwell.
-*     Jeremy Du Croz, Numerical Algorithms Group Ltd.
-*     Sven Hammarling, Numerical Algorithms Group Ltd.
+*  =====================================================================
 *
 *     .. Parameters ..
       INTEGER            NIN
@@ -54,8 +101,8 @@
       COMPLEX*16         ZERO, ONE
       PARAMETER          ( ZERO = ( 0.0D0, 0.0D0 ),
      $                   ONE = ( 1.0D0, 0.0D0 ) )
-      DOUBLE PRECISION   RZERO, RHALF, RONE
-      PARAMETER          ( RZERO = 0.0D0, RHALF = 0.5D0, RONE = 1.0D0 )
+      DOUBLE PRECISION   RZERO
+      PARAMETER          ( RZERO = 0.0D0 )
       INTEGER            NMAX
       PARAMETER          ( NMAX = 65 )
       INTEGER            NIDMAX, NALMAX, NBEMAX
@@ -104,7 +151,7 @@
 *
       READ( NIN, FMT = * )SUMMRY
       READ( NIN, FMT = * )NOUT
-      OPEN( NOUT, FILE = SUMMRY, STATUS = 'NEW' )
+      OPEN( NOUT, FILE = SUMMRY, STATUS = 'UNKNOWN' )
       NOUTC = NOUT
 *
 *     Read name and unit number for snapshot output file and open file.
@@ -113,7 +160,7 @@
       READ( NIN, FMT = * )NTRA
       TRACE = NTRA.GE.0
       IF( TRACE )THEN
-         OPEN( NTRA, FILE = SNAPS, STATUS = 'NEW' )
+         OPEN( NTRA, FILE = SNAPS, STATUS = 'UNKNOWN' )
       END IF
 *     Read the flag that directs rewinding of the snapshot file.
       READ( NIN, FMT = * )REWI
@@ -190,14 +237,7 @@
 *
 *     Compute EPS (the machine precision).
 *
-      EPS = RONE
-   70 CONTINUE
-      IF( DDIFF( RONE + EPS, RONE ).EQ.RZERO )
-     $   GO TO 80
-      EPS = RHALF*EPS
-      GO TO 70
-   80 CONTINUE
-      EPS = EPS + EPS
+      EPS = EPSILON(RZERO)
       WRITE( NOUT, FMT = 9998 )EPS
 *
 *     Check the reliability of ZMMCH using exact data.
@@ -1303,8 +1343,6 @@
       NC = 0
       RESET = .TRUE.
       ERRMAX = RZERO
-      RALS = RONE
-      RBETS = RONE
 *
       DO 100 IN = 1, NIDIM
          N = IDIM( IN )
@@ -1951,7 +1989,7 @@
 *
 *  Tests the error exits from the Level 3 Blas.
 *  Requires a special version of the error-handling routine XERBLA.
-*  ALPHA, RALPHA, BETA, RBETA, A, B and C should not need to be defined.
+*  A, B and C should not need to be defined.
 *
 *  Auxiliary routine for test program for Level 3 Blas.
 *
@@ -1961,12 +1999,20 @@
 *     Jeremy Du Croz, Numerical Algorithms Group Ltd.
 *     Sven Hammarling, Numerical Algorithms Group Ltd.
 *
+*  3-19-92:  Initialize ALPHA, BETA, RALPHA, and RBETA  (eca)
+*  3-19-92:  Fix argument 12 in calls to ZSYMM and ZHEMM
+*            with INFOT = 9  (eca)
+*  10-9-00:  Declared INTRINSIC DCMPLX (susan)
+*
 *     .. Scalar Arguments ..
       INTEGER            ISNUM, NOUT
       CHARACTER*6        SRNAMT
 *     .. Scalars in Common ..
       INTEGER            INFOT, NOUTC
       LOGICAL            LERR, OK
+*     .. Parameters ..
+      REAL               ONE, TWO
+      PARAMETER          ( ONE = 1.0D0, TWO = 2.0D0 )
 *     .. Local Scalars ..
       COMPLEX*16         ALPHA, BETA
       DOUBLE PRECISION   RALPHA, RBETA
@@ -1975,6 +2021,8 @@
 *     .. External Subroutines ..
       EXTERNAL           ZGEMM, ZHEMM, ZHER2K, ZHERK, CHKXER, ZSYMM,
      $                   ZSYR2K, ZSYRK, ZTRMM, ZTRSM
+*     .. Intrinsic Functions ..
+      INTRINSIC          DCMPLX
 *     .. Common blocks ..
       COMMON             /INFOC/INFOT, NOUTC, OK, LERR
 *     .. Executable Statements ..
@@ -1984,6 +2032,14 @@
 *     LERR is set to .TRUE. by the special version of XERBLA each time
 *     it is called, and is then tested and re-set by CHKXER.
       LERR = .FALSE.
+*
+*     Initialize ALPHA, BETA, RALPHA, and RBETA.
+*
+      ALPHA = DCMPLX( ONE, -ONE )
+      BETA = DCMPLX( TWO, -TWO )
+      RALPHA = ONE
+      RBETA = TWO
+*
       GO TO ( 10, 20, 30, 40, 50, 60, 70, 80,
      $        90 )ISNUM
    10 INFOT = 1
@@ -2210,16 +2266,16 @@
       CALL ZHEMM( 'R', 'L', 0, 2, ALPHA, A, 1, B, 1, BETA, C, 1 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL ZHEMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 1 )
+      CALL ZHEMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL ZHEMM( 'R', 'U', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 1 )
+      CALL ZHEMM( 'R', 'U', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL ZHEMM( 'L', 'L', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 1 )
+      CALL ZHEMM( 'L', 'L', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL ZHEMM( 'R', 'L', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 1 )
+      CALL ZHEMM( 'R', 'L', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 12
       CALL ZHEMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 2, BETA, C, 1 )
@@ -2277,16 +2333,16 @@
       CALL ZSYMM( 'R', 'L', 0, 2, ALPHA, A, 1, B, 1, BETA, C, 1 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL ZSYMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 1 )
+      CALL ZSYMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL ZSYMM( 'R', 'U', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 1 )
+      CALL ZSYMM( 'R', 'U', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL ZSYMM( 'L', 'L', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 1 )
+      CALL ZSYMM( 'L', 'L', 2, 0, ALPHA, A, 2, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 9
-      CALL ZSYMM( 'R', 'L', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 1 )
+      CALL ZSYMM( 'R', 'L', 2, 0, ALPHA, A, 1, B, 1, BETA, C, 2 )
       CALL CHKXER( SRNAMT, INFOT, NOUT, LERR, OK )
       INFOT = 12
       CALL ZSYMM( 'L', 'U', 2, 0, ALPHA, A, 2, B, 2, BETA, C, 1 )
@@ -3276,7 +3332,6 @@
    50    CONTINUE
       END IF
 *
-   60 CONTINUE
       LZERES = .TRUE.
       GO TO 80
    70 CONTINUE
diff --git a/utest/CMakeLists.txt b/utest/CMakeLists.txt
index 544646911..dc5175fc5 100644
--- a/utest/CMakeLists.txt
+++ b/utest/CMakeLists.txt
@@ -6,6 +6,7 @@ if (MSVC AND "${CMAKE_C_COMPILER_ID}" MATCHES Clang)
 else ()
   set(OpenBLAS_utest_src
     utest_main.c
+    test_min.c
     test_amax.c
     test_ismin.c
     test_rotmg.c
diff --git a/utest/Makefile b/utest/Makefile
index 32bdcc6e1..0b9892411 100644
--- a/utest/Makefile
+++ b/utest/Makefile
@@ -11,7 +11,7 @@ UTESTBIN=openblas_utest
 
 include $(TOPDIR)/Makefile.system
 
-OBJS=utest_main.o test_amax.o test_ismin.o test_rotmg.o test_axpy.o test_dotu.o test_dsdot.o test_swap.o test_rot.o
+OBJS=utest_main.o test_min.o test_amax.o test_ismin.o test_rotmg.o test_axpy.o test_dotu.o test_dsdot.o test_swap.o test_rot.o
 #test_rot.o test_swap.o test_axpy.o test_dotu.o test_dsdot.o test_fork.o
 
 ifneq ($(NO_LAPACK), 1)
diff --git a/utest/test_amax.c b/utest/test_amax.c
index 411598410..831804027 100644
--- a/utest/test_amax.c
+++ b/utest/test_amax.c
@@ -40,6 +40,17 @@ CTEST(amax, samax){
 
   te_max=BLASFUNC(samax)(&N, x, &inc);
   tr_max=3.3;
-  
+
   ASSERT_DBL_NEAR_TOL((double)(tr_max), (double)(te_max), SINGLE_EPS);
 }
+
+CTEST(amax, damax){
+  blasint N=3, inc=1;
+  double te_max=0.0, tr_max=0.0;
+  double x[]={-1.1, 2.2, -3.3};
+
+  te_max=BLASFUNC(damax)(&N, x, &inc);
+  tr_max=3.3;
+
+  ASSERT_DBL_NEAR_TOL((double)(tr_max), (double)(te_max), DOUBLE_EPS);
+}
diff --git a/utest/test_min.c b/utest/test_min.c
new file mode 100644
index 000000000..fd31b5982
--- /dev/null
+++ b/utest/test_min.c
@@ -0,0 +1,100 @@
+/*****************************************************************************
+Copyright (c) 2011-2016, The OpenBLAS Project
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+
+   1. Redistributions of source code must retain the above copyright
+      notice, this list of conditions and the following disclaimer.
+
+   2. Redistributions in binary form must reproduce the above copyright
+      notice, this list of conditions and the following disclaimer in
+      the documentation and/or other materials provided with the
+      distribution.
+   3. Neither the name of the OpenBLAS project nor the names of
+      its contributors may be used to endorse or promote products
+      derived from this software without specific prior written
+      permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+**********************************************************************************/
+
+#include "openblas_utest.h"
+
+CTEST(min, smin_negative){
+  blasint N=3, inc=1;
+  float te_min=0.0, tr_min=0.0;
+  float x[]={-1.1, -2.2, -3.3};
+
+  te_min=BLASFUNC(smin)(&N, x, &inc);
+  tr_min=-3.3;
+
+  ASSERT_DBL_NEAR_TOL((double)(tr_min), (double)(te_min), SINGLE_EPS);
+}
+
+CTEST(min, dmin_positive){
+  blasint N=3, inc=1;
+  double te_min=0.0, tr_min=0.0;
+  double x[]={1.1, 0.0, 3.3};
+
+  te_min=BLASFUNC(dmin)(&N, x, &inc);
+  tr_min=0.0;
+
+  ASSERT_DBL_NEAR_TOL((double)(tr_min), (double)(te_min), DOUBLE_EPS);
+}
+
+CTEST(min, smin_zero){
+  blasint N=3, inc=1;
+  float te_min=0.0, tr_min=0.0;
+  float x[]={1.1, 2.2, 0.0};
+
+  te_min=BLASFUNC(smin)(&N, x, &inc);
+  tr_min=0.0;
+
+  ASSERT_DBL_NEAR_TOL((double)(tr_min), (double)(te_min), SINGLE_EPS);
+}
+
+CTEST(max, smax_negative){
+  blasint N=3, inc=1;
+  float te_max=0.0, tr_max=0.0;
+  float x[]={-1.1, -2.2, -3.3};
+
+  te_max=BLASFUNC(smax)(&N, x, &inc);
+  tr_max=-1.1;
+
+  ASSERT_DBL_NEAR_TOL((double)(tr_max), (double)(te_max), SINGLE_EPS);
+}
+
+CTEST(max, dmax_positive){
+  blasint N=3, inc=1;
+  double te_max=0.0, tr_max=0.0;
+  double x[]={1.1, 0.0, 3.3};
+
+  te_max=BLASFUNC(dmax)(&N, x, &inc);
+  tr_max=3.3;
+
+  ASSERT_DBL_NEAR_TOL((double)(tr_max), (double)(te_max), DOUBLE_EPS);
+}
+
+CTEST(max, smax_zero){
+  blasint N=3, inc=1;
+  float te_max=0.0, tr_max=0.0;
+  float x[]={-1.1, -2.2, 0.0};
+
+  te_max=BLASFUNC(smax)(&N, x, &inc);
+  tr_max=0.0;
+
+  ASSERT_DBL_NEAR_TOL((double)(tr_max), (double)(te_max), SINGLE_EPS);
+}
diff --git a/utest/utest_main2.c b/utest/utest_main2.c
index aa95a5a3f..6b252863a 100644
--- a/utest/utest_main2.c
+++ b/utest/utest_main2.c
@@ -50,6 +50,17 @@ CTEST(amax, samax){
   ASSERT_DBL_NEAR_TOL((double)(tr_max), (double)(te_max), SINGLE_EPS);
 }
 
+CTEST(amax, damax){
+  blasint N=3, inc=1;
+  double te_max=0.0, tr_max=0.0;
+  double x[]={-1.1, 2.2, -3.3};
+
+  te_max=BLASFUNC(damax)(&N, x, &inc);
+  tr_max=3.3;
+
+  ASSERT_DBL_NEAR_TOL((double)(tr_max), (double)(te_max), DOUBLE_EPS);
+}
+
 CTEST (drotmg,rotmg)
 {
 	double te_d1, tr_d1;
@@ -508,9 +519,82 @@ CTEST(swap,cswap_inc_0)
 	}
 }
 
+CTEST(min, smin_negative){
+  blasint N=3, inc=1;
+  float te_min=0.0, tr_min=0.0;
+  float x[]={-1.1, -2.2, -3.3};
+
+  te_min=BLASFUNC(smin)(&N, x, &inc);
+  tr_min=-3.3;
+
+  ASSERT_DBL_NEAR_TOL((double)(tr_min), (double)(te_min), SINGLE_EPS);
+}
+
+CTEST(min, dmin_positive){
+  blasint N=3, inc=1;
+  double te_min=0.0, tr_min=0.0;
+  double x[]={1.1, 0.0, 3.3};
+
+  te_min=BLASFUNC(dmin)(&N, x, &inc);
+  tr_min=0.0;
+
+  ASSERT_DBL_NEAR_TOL((double)(tr_min), (double)(te_min), DOUBLE_EPS);
+}
+
+CTEST(min, smin_zero){
+  blasint N=3, inc=1;
+  float te_min=0.0, tr_min=0.0;
+  float x[]={1.1, 2.2, 0.0};
+
+  te_min=BLASFUNC(smin)(&N, x, &inc);
+  tr_min=0.0;
+
+  ASSERT_DBL_NEAR_TOL((double)(tr_min), (double)(te_min), SINGLE_EPS);
+}
+
+CTEST(max, smax_negative){
+  blasint N=3, inc=1;
+  float te_max=0.0, tr_max=0.0;
+  float x[]={-1.1, -2.2, -3.3};
+
+  te_max=BLASFUNC(smax)(&N, x, &inc);
+  tr_max=-1.1;
+
+  ASSERT_DBL_NEAR_TOL((double)(tr_max), (double)(te_max), SINGLE_EPS);
+}
+
+CTEST(max, dmax_positive){
+  blasint N=3, inc=1;
+  double te_max=0.0, tr_max=0.0;
+  double x[]={1.1, 0.0, 3.3};
+
+  te_max=BLASFUNC(dmax)(&N, x, &inc);
+  tr_max=3.3;
+
+  ASSERT_DBL_NEAR_TOL((double)(tr_max), (double)(te_max), DOUBLE_EPS);
+}
+
+CTEST(max, smax_zero){
+  blasint N=3, inc=1;
+  float te_max=0.0, tr_max=0.0;
+  float x[]={-1.1, -2.2, 0.0};
+
+  te_max=BLASFUNC(smax)(&N, x, &inc);
+  tr_max=0.0;
+
+  ASSERT_DBL_NEAR_TOL((double)(tr_max), (double)(te_max), SINGLE_EPS);
+}
+
 int main(int argc, const char ** argv){
 
-  CTEST_ADD(amax, samax);
+  CTEST_ADD (amax, samax);
+  CTEST_ADD (amax, damax);
+  CTEST_ADD (min, smin_negative); 
+  CTEST_ADD (min, dmin_positive); 
+  CTEST_ADD (min, smin_zero); 
+  CTEST_ADD (max, smax_negative); 
+  CTEST_ADD (max, dmax_positive); 
+  CTEST_ADD (max, smax_zero); 
   CTEST_ADD (drotmg,rotmg);
   CTEST_ADD (drotmg,rotmg_issue1452);
   CTEST_ADD (drotmg,rotmg_D1eqD2_X1eqX2);